# EvalMap · one brief, 142 AI-built 3D jeep race games

Source page: https://evalmap.ai/jeep/

The jeep gallery · 16 Jul 2026 – 7 Nov 2026

# One brief, 142 builds.

The same brief — a drivable 3D jeep race that runs in a browser page — went to a series of models and agents between 16 Jul 2026 and 7 Nov 2026. Every answer was published with only the edits listed below, and every one that works runs here: press play and drive it, no download, nothing to install. 41 produced no game at all; their cards say why, with a drawing where the frame would be.

One brief, and a board of EvalMap’s own: every build that runs is scored 0–100 against a written rubric, where 100 is a game on the level of GTA VI. 97 models are named, 24 of them wrote more than one build, and 9 builds record no model. The scores stay here: they feed nothing on the atlas, and the atlas measures none of them.

[Play #1: Pareto 26.10 Preview 58/100 ▶](https://evalmap.ai/jeep/pareto-2610-preview/)[The board ↓](https://evalmap.ai/jeep/#jeep-board-title)[Every build ↓](https://evalmap.ai/jeep/#jeep-index-title)

142builds83scored41failed the brief51also on the atlas

Pause the frames

Top of the board

## The highest scores so far

Every build ranked third or better, ties included, on a scale where 100 is a game on the level of GTA VI. Each frame opens the game.

- [A frame from the jeep race written by Pareto 26.10 Preview1Play it ▶](https://evalmap.ai/jeep/pareto-2610-preview/)

58/100

a solid hobby browser game

Ranked 1: [Pareto 26.10 Preview](https://evalmap.ai/jeep/#pareto-2610-preview)

Ridge Runner — Alpine Expedition

- [A frame from the jeep race written by Claude Sonnet 5.52Play it ▶](https://evalmap.ai/jeep/sonnet55-xhigh/)

48/100

a solid hobby browser game

Ranked 2: [Claude Sonnet 5.5](https://evalmap.ai/jeep/#sonnet55-xhigh) xhigh effort

Highlands Rally — Jeep Race

- [A frame from the jeep race written by Mistral Large 4 Preview2Play it ▶](https://evalmap.ai/jeep/mistral-large4-preview/)

48/100

a solid hobby browser game

Ranked 2: [Mistral Large 4 Preview](https://evalmap.ai/jeep/#mistral-large4-preview)

Red Rock Rally

83 builds scored

## The board

Completed build reviews score 0–100 against the same rubric, where 100 is a finished open-world driving game on the level of GTA VI. On that ladder 20 is a rudimentary demo, 40 a solid hobby browser game and 60 a game a studio could sell — so a build written from one line of brief belongs in the thirties if it plays. Each is first driven for a minute, and straight on from the start for 90 seconds to find the edge of its world; a build whose jeep flips or wedges itself on the course is held to 20, and one whose jeep falls out of the world, drives off its edge into nothing or comes apart to 12. What plays is scored, not what the source promises. The score is EvalMap’s own judgement of the build as published here, in six parts on each card, and it says nothing about the model beyond this one game.

- 0–19less than a rudimentary demo
- 20–39a rudimentary demo
- 40–59a solid hobby browser game
- 60–79a good commercial indie game
- 80–99an AA production
- 100GTA VI

Best so far 58

- [1Pareto 26.10 Preview58](https://evalmap.ai/jeep/#pareto-2610-preview)
- [2Claude Sonnet 5.5 xhigh effort48](https://evalmap.ai/jeep/#sonnet55-xhigh)
- [2Mistral Large 4 Preview48](https://evalmap.ai/jeep/#mistral-large4-preview)
- [4MiMo 2.6 Flash46](https://evalmap.ai/jeep/#mimo-v26-flash-fb)
- [5Space Bunny Alpha max effort45](https://evalmap.ai/jeep/#space-bunny-alpha-max-zen-r2)
- [5GPT-6.1 Sol xhigh effort45](https://evalmap.ai/jeep/#gpt61sol-xhigh)
- [7Qwen3.8 Max Prime max effort42](https://evalmap.ai/jeep/#qwen38-max-prime)
- [7MiMo-V2.6-Pro42](https://evalmap.ai/jeep/#mimo-v26-pro-fb-r2)
- [9GPT-5.6 Sol max effort41](https://evalmap.ai/jeep/#gpt56-sol-max-cmd)
- [10Claude Opus 5.5 xhigh effort40](https://evalmap.ai/jeep/#opus55-xhigh-cc)
- [10DeepSeek V4.1 Flash max effort40](https://evalmap.ai/jeep/#deepseek-v41-flash-max-fb)
- [10Grok 4.7 xhigh effort40](https://evalmap.ai/jeep/#grok47-build-fast-xhigh)
- [13GPT-6 Astra39](https://evalmap.ai/jeep/#astra)
- [13GPT-6 Sol high effort39](https://evalmap.ai/jeep/#gpt6sol-high-cx)
- [13Qwen3.8-Flash-Next on the bench39](https://evalmap.ai/jeep/#qwen38-flash-oc2)
- [13Claude Opus 5 max effort39](https://evalmap.ai/jeep/#claude-opus-5-max-cmd)
- [17Muse Spark 1.338](https://evalmap.ai/jeep/#muse13)
- [18MiMo-X-Pro-Preview37](https://evalmap.ai/jeep/#mimo-x-pro)
- [18Ling-3.1-flash high effort37](https://evalmap.ai/jeep/#ling31-flash-high-oc2)
- [20MiMo-V2.6-Pro36](https://evalmap.ai/jeep/#mimo-v26-pro)
- [20Space Bunny Alpha max effort36](https://evalmap.ai/jeep/#space-bunny-alpha-max)
- [20GPT-5.6 Luna max effort36](https://evalmap.ai/jeep/#gpt56-luna-max-cmd)
- [23HY4 Preview35](https://evalmap.ai/jeep/#tecent-hy4-preview)
- [23Muse Spark 1.3 max effort35](https://evalmap.ai/jeep/#muse13-max-effort)
- [23Grok 4.7 xhigh effort35](https://evalmap.ai/jeep/#grok47-xhigh)
- [26Pareto 26.934](https://evalmap.ai/jeep/#union-alpha-092026)
- [26DeepSeek-V4-Flash-Vision-Exp max effort34](https://evalmap.ai/jeep/#deepseek-v4-flash-vision-exp-max-oc2)
- [28GLM-5.3-Flash on two DGX Sparks33](https://evalmap.ai/jeep/#glm53-flash-2x-sparks-localai)
- [28GLM-5.3 max effort33](https://evalmap.ai/jeep/#glm53-max-oc2)
- [28Kimi K3 max effort33](https://evalmap.ai/jeep/#kimi-k3-max-cmd)
- [28GPT-5.6 Terra max effort33](https://evalmap.ai/jeep/#gpt56-terra-max-cmd)
- [32Jeep Race model unknown32](https://evalmap.ai/jeep/#jeep-race)
- [32GLM-5.332](https://evalmap.ai/jeep/#glm53)
- [32Claude Fable 5.132](https://evalmap.ai/jeep/#fable51)
- [32Qwen3.7 Max on the bench32](https://evalmap.ai/jeep/#qwen37-cmd)
- [32LongCat-2.5-Preview on the bench32](https://evalmap.ai/jeep/#longcat25-preview-oc2)
- [32Hy3 high effort32](https://evalmap.ai/jeep/#hy3-oc2-r2)
- [38GLM-5.3-Flash31](https://evalmap.ai/jeep/#glm5-3-flash)
- [38Claude Fable 5.1 medium effort31](https://evalmap.ai/jeep/#fable51-medium-effort)
- [38Devin SWE-231](https://evalmap.ai/jeep/#devin-swe2)
- [38Step 5 Preview31](https://evalmap.ai/jeep/#step-5-preview-stepfun)
- [42Muse Spark 1.230](https://evalmap.ai/jeep/#muse12)
- [43Jeep Road Runner model unknown29](https://evalmap.ai/jeep/#jeep-road-runner)
- [43Qwen3.8 Flash on two DGX Sparks29](https://evalmap.ai/jeep/#qwen38flash-local-2xsparks)
- [45Jeep Racing model unknown28](https://evalmap.ai/jeep/#jeep-rally-webgpu)
- [45Step 5 Preview (high)28](https://evalmap.ai/jeep/#step5-preview-high)
- [47Jeep Rally model unknown27](https://evalmap.ai/jeep/#jeep-rally)
- [47Qwen3.8 27B on an RTX 409027](https://evalmap.ai/jeep/#qwen38-27b-local-4090rtx)
- [49Desert Rally model unknown25](https://evalmap.ai/jeep/#jeep-racing)
- [50Jeep Race model unknown24](https://evalmap.ai/jeep/#jeep-race-webgpu)
- [50Qwen3.8 Max 090224](https://evalmap.ai/jeep/#qwen38-max-0902)
- [52DESERT RAID model unknown23](https://evalmap.ai/jeep/#desert-jeep-rally)
- [52Ling-3.0-flash-VL on the bench23](https://evalmap.ai/jeep/#ling3-flash-vl-oc2)
- [54MiniMax-M2.7 on the bench22](https://evalmap.ai/jeep/#minimax-m27-oc2)
- [55DUST RAIDERS model unknown20](https://evalmap.ai/jeep/#dust-raiders)
- [56MudTrail 4x4 model unknown12](https://evalmap.ai/jeep/#jeep-webgpu-race)
- [56Gemini 3.7 Flash12](https://evalmap.ai/jeep/#gemini-3-7-flash)
- [56Gemini 3.8 Flash12](https://evalmap.ai/jeep/#gemini38flash)
- [56Gemini 3.8 Flash second build12](https://evalmap.ai/jeep/#aaa-jeep-game-gemini38flash)
- [56Omen Alpha12](https://evalmap.ai/jeep/#omen-alpha-maybe-glm5)
- [56Qwen3.8 27B on Cerebras12](https://evalmap.ai/jeep/#cerebrast-qwen38-27b-1300toksec-medium-effort)
- [56DeepSeek V4.1 Flash12](https://evalmap.ai/jeep/#deepseek-v4-1-flash-beta)
- [56Kimi K2.8 Preview12](https://evalmap.ai/jeep/#kimi-k28preview)
- [56GLM-5.3-FlashX12](https://evalmap.ai/jeep/#glm53-flashx)
- [56Fugu Max12](https://evalmap.ai/jeep/#fugu-max)
- [56GPT-6 Luna max effort12](https://evalmap.ai/jeep/#gpt6luna-max-cx)
- [56Claude Sonnet 5 max effort12](https://evalmap.ai/jeep/#claude-sonnet-5-max-cmd)
- [56Inkling on the bench12](https://evalmap.ai/jeep/#inkling-cmd)
- [56Gemini 3.6 Flash default effort12](https://evalmap.ai/jeep/#gemini36-flash-cmd)
- [56Gemini 3.8 Flash high effort12](https://evalmap.ai/jeep/#gemini38-flash-high-cmd)
- [56Qwen3.7 Plus on the bench12](https://evalmap.ai/jeep/#qwen37-plus-cmd)
- [56Gemini 3.6 Flash high effort12](https://evalmap.ai/jeep/#gemini36-flash-high-cmd)
- [56MiMo-V2.5 high effort12](https://evalmap.ai/jeep/#mimo-v25-high-oc2)
- [56Qwen3 Coder Next on the bench12](https://evalmap.ai/jeep/#qwen3-coder-next-oc2)
- [56GLM-5.1 high effort12](https://evalmap.ai/jeep/#glm51-high-oc2)
- [56GPT-6 Luna max effort12](https://evalmap.ai/jeep/#gpt6luna-max-fb)
- [56MiniMax-M3.1-Flash-Preview12](https://evalmap.ai/jeep/#minimax-m31-flash-preview-max-mcode)
- [56Gemini 3.5 Flash high effort12](https://evalmap.ai/jeep/#gemini35-flash-high-cmd-r2)
- [56GLM-5.3-Flash max effort12](https://evalmap.ai/jeep/#glm53-flash-max-fb-r2)
- [56GLM-5.3-Prime max effort12](https://evalmap.ai/jeep/#glm53-prime-max-oc2-r2)
- [56Gemini 3.8 Flash high effort12](https://evalmap.ai/jeep/#gemini38-flash-high-agy)
- [56Pixel Canary xhigh effort12](https://evalmap.ai/jeep/#pixel-canary-xhigh)
- [56DeepSeek V4.1 Flash 101712](https://evalmap.ai/jeep/#deepseek-v41-flash-1017)

Not scored: [Claude Mythos 5.1 (xhigh)](https://evalmap.ai/jeep/#mythos51-xhigh), [Claude Fable 5.1 (xhigh)](https://evalmap.ai/jeep/#fable51-xhigh), [Claude Haiku 5.5](https://evalmap.ai/jeep/#haiku55), [GLM-5.3-Prime](https://evalmap.ai/jeep/#glm53-prime-g22) — their scoring review is incomplete. Each card explains why.

Failed the brief: [AliceAI-Foundation-80B-A3B-Base](https://evalmap.ai/jeep/#aliceai-foundation-80b), [Solar Mini 4](https://evalmap.ai/jeep/#solar-mini4-max-oc2), [Mercury 2.5](https://evalmap.ai/jeep/#mercury25-high-oc2), [Granite 4.2 8B](https://evalmap.ai/jeep/#granite42-8b-high-oc2), [Ling-3.0-flash-Fin](https://evalmap.ai/jeep/#ling3-flash-fin-oc2), [GLM-5.3-Prime](https://evalmap.ai/jeep/#glm53-prime-max-oc2), [DeepSeek V4 Pro 0813](https://evalmap.ai/jeep/#deepseek-v4-pro-0813-oc2), [MiniMax-M3](https://evalmap.ai/jeep/#minimax-m3-oc2), [GLM-5.2](https://evalmap.ai/jeep/#glm52-oc2), [Qwen3.5 397B A17B](https://evalmap.ai/jeep/#qwen35-397b-oc2), [DeepSeek V4 Flash 0731](https://evalmap.ai/jeep/#deepseek-v4-flash-0731-oc2), [Ling-3.0-flash](https://evalmap.ai/jeep/#ling3-flash-oc2), [Kimi K2.6](https://evalmap.ai/jeep/#kimi-k26-oc2), [Hy3](https://evalmap.ai/jeep/#hy3-oc2), [MiMo-V2.5-Pro](https://evalmap.ai/jeep/#mimo-v25-pro-oc2), [Kimi K2.7 Code](https://evalmap.ai/jeep/#kimi-k27-code-oc2), [Inkling Small](https://evalmap.ai/jeep/#inkling-small-cmd), [Gemini 3.5 Flash](https://evalmap.ai/jeep/#gemini35-flash-high-cmd), [Nemotron 3.5 Lightning](https://evalmap.ai/jeep/#nemotron35-lightning-or), [Gemma 4 31B](https://evalmap.ai/jeep/#gemma4-31b-oc2), [GLM-5.3-Flash](https://evalmap.ai/jeep/#glm53-flash-max-fb), [Solar Mini 4](https://evalmap.ai/jeep/#solar-mini4-fb), [Solar Pro 4](https://evalmap.ai/jeep/#solar-pro4-fb), [MiMo-V2.6-Pro](https://evalmap.ai/jeep/#mimo-v26-pro-fb), [Gemini 3.8 Flash](https://evalmap.ai/jeep/#gemini38-flash-max-fb), [gpt-oss-120b](https://evalmap.ai/jeep/#gpt-oss-120b-high-oc2), [Space Bunny Alpha](https://evalmap.ai/jeep/#space-bunny-alpha-max-zen), [Gemini 3.8 Flash](https://evalmap.ai/jeep/#gemini38-flash-max-fb-r2), [Kimi K2.6](https://evalmap.ai/jeep/#kimi-k26-oc2-r2), [Mercury 2.5](https://evalmap.ai/jeep/#mercury25-high-oc2-r2), [Granite 4.2 8B](https://evalmap.ai/jeep/#granite42-8b-high-oc2-r2), [Solar Pro 4](https://evalmap.ai/jeep/#solar-pro4-fb-r2), [Ling-3.0-flash-Fin](https://evalmap.ai/jeep/#ling3-flash-fin-oc2-r2), [Nemotron 3.5 Lightning](https://evalmap.ai/jeep/#nemotron35-lightning-or-r2), [Solar Mini 4](https://evalmap.ai/jeep/#solar-mini4-max-oc2-r2), [DeepSeek V4 Pro 0813](https://evalmap.ai/jeep/#deepseek-v4-pro-0813-oc2-r2), [Solar Mini 4](https://evalmap.ai/jeep/#solar-mini4-fb-r2), [Fledge Alpha Free](https://evalmap.ai/jeep/#fledge-alpha-oc2), [Grok 4.7](https://evalmap.ai/jeep/#grok47-xhigh-oc2), [Kolibri-1](https://evalmap.ai/jeep/#kolibri-1-oc2), [Nemotron 3.5 Super VL](https://evalmap.ai/jeep/#nemotron35-super-vl) — no game to drive, so no score. Each card says why.

All 142, newest first

## Jump to a build

3–7 Nov 2026

- [Qoduck (Qwen4 Flash/Plus/Max?) (xhigh)7 Nov](https://evalmap.ai/jeep/#qoduck-xhigh)
- [Mercury 36 Nov](https://evalmap.ai/jeep/#mercury3-high)
- [Devin SWE-2 Flash (max)6 Nov](https://evalmap.ai/jeep/#devin-swe2-flash-max)
- [LongCat 2.5 (max)3 Nov](https://evalmap.ai/jeep/#longcat25-max)

27 Oct

- [GPT-6.1 Terra (max)27 Oct](https://evalmap.ai/jeep/#gpt61terra-max)

19–23 Oct

- [Kimi K3.1 Pro (max)23 Oct](https://evalmap.ai/jeep/#kimi-k31-pro-max)
- [Kimi K2.7 Codesecond build19 Oct](https://evalmap.ai/jeep/#kimi-k27-code)
- [MiniMax-M3.1-Flash19 Oct](https://evalmap.ai/jeep/#minimax-m31-flash)

12–18 Oct

- [Beam 1.0 (max)18 Oct](https://evalmap.ai/jeep/#reflection-beam-max)
- [DeepSeek V4.1 Flash 101718 Oct](https://evalmap.ai/jeep/#deepseek-v41-flash-1017)
- [Nemotron 3.5 Super VLfailed15 Oct](https://evalmap.ai/jeep/#nemotron35-super-vl)
- [Gemini 4 Argon (max, cli2027)14 Oct](https://evalmap.ai/jeep/#gemini4-argon-max-cli2027)
- [Gemini 4 Argon (max)14 Oct](https://evalmap.ai/jeep/#gemini4-argon-max)
- [Devin SWE-2 (max)13 Oct](https://evalmap.ai/jeep/#devin-swe2-max)
- [Gemini 4 Argon (high)13 Oct](https://evalmap.ai/jeep/#gemini4-argon-high)
- [Muse Wonder Full (max)12 Oct](https://evalmap.ai/jeep/#muse-wonder-full-max)

5–8 Oct

- [GLM-5.3-Prime8 Oct](https://evalmap.ai/jeep/#glm53-prime-g22)
- [Step 5 Preview (high)8 Oct](https://evalmap.ai/jeep/#step5-preview-high)
- [Claude Haiku 5.57 Oct](https://evalmap.ai/jeep/#haiku55)
- [Claude Fable 5.1 (xhigh)7 Oct](https://evalmap.ai/jeep/#fable51-xhigh)
- [Claude Mythos 5.1 (xhigh)7 Oct](https://evalmap.ai/jeep/#mythos51-xhigh)
- [Mistral Large 4 Preview6 Oct](https://evalmap.ai/jeep/#mistral-large4-preview)
- [Kolibri-1failed5 Oct](https://evalmap.ai/jeep/#kolibri-1-oc2)

28 Sep–3 Oct

- [Pareto 26.10 Preview3 Oct](https://evalmap.ai/jeep/#pareto-2610-preview)
- [Grok 4.7xhigh effort2 Oct](https://evalmap.ai/jeep/#grok47-build-fast-xhigh)
- [Grok 4.7failed2 Oct](https://evalmap.ai/jeep/#grok47-xhigh-oc2)
- [Fledge Alpha Freefailed2 Oct](https://evalmap.ai/jeep/#fledge-alpha-oc2)
- [Ling-3.1-flashhigh effort30 Sep](https://evalmap.ai/jeep/#ling31-flash-high-oc2)
- [GPT-6.1 Solxhigh effort29 Sep](https://evalmap.ai/jeep/#gpt61sol-xhigh)
- [Claude Sonnet 5.5xhigh effort28 Sep](https://evalmap.ai/jeep/#sonnet55-xhigh)
- [Pixel Canaryxhigh effort28 Sep](https://evalmap.ai/jeep/#pixel-canary-xhigh)

21–27 Sep

- [Solar Mini 4failed27 Sep](https://evalmap.ai/jeep/#solar-mini4-fb-r2)
- [DeepSeek V4 Pro 0813failed27 Sep](https://evalmap.ai/jeep/#deepseek-v4-pro-0813-oc2-r2)
- [Solar Mini 4failed27 Sep](https://evalmap.ai/jeep/#solar-mini4-max-oc2-r2)
- [Nemotron 3.5 Lightningfailed27 Sep](https://evalmap.ai/jeep/#nemotron35-lightning-or-r2)
- [Ling-3.0-flash-Finfailed27 Sep](https://evalmap.ai/jeep/#ling3-flash-fin-oc2-r2)
- [Solar Pro 4failed27 Sep](https://evalmap.ai/jeep/#solar-pro4-fb-r2)
- [Granite 4.2 8Bfailed27 Sep](https://evalmap.ai/jeep/#granite42-8b-high-oc2-r2)
- [Mercury 2.5failed27 Sep](https://evalmap.ai/jeep/#mercury25-high-oc2-r2)
- [Kimi K2.6failed27 Sep](https://evalmap.ai/jeep/#kimi-k26-oc2-r2)
- [Gemini 3.8 Flashfailed27 Sep](https://evalmap.ai/jeep/#gemini38-flash-max-fb-r2)
- [Space Bunny Alphafailed27 Sep](https://evalmap.ai/jeep/#space-bunny-alpha-max-zen)
- [Gemini 3.8 Flashhigh effort27 Sep](https://evalmap.ai/jeep/#gemini38-flash-high-agy)
- [Hy3high effort27 Sep](https://evalmap.ai/jeep/#hy3-oc2-r2)
- [GLM-5.3-Primemax effort27 Sep](https://evalmap.ai/jeep/#glm53-prime-max-oc2-r2)
- [Space Bunny Alphamax effort27 Sep](https://evalmap.ai/jeep/#space-bunny-alpha-max-zen-r2)
- [GLM-5.3-Flashmax effort27 Sep](https://evalmap.ai/jeep/#glm53-flash-max-fb-r2)
- [MiMo-V2.6-Pro27 Sep](https://evalmap.ai/jeep/#mimo-v26-pro-fb-r2)
- [Gemini 3.5 Flashhigh effort27 Sep](https://evalmap.ai/jeep/#gemini35-flash-high-cmd-r2)
- [MiniMax-M3.1-Flash-Preview27 Sep](https://evalmap.ai/jeep/#minimax-m31-flash-preview-max-mcode)
- [gpt-oss-120bfailed26 Sep](https://evalmap.ai/jeep/#gpt-oss-120b-high-oc2)
- [Gemini 3.8 Flashfailed26 Sep](https://evalmap.ai/jeep/#gemini38-flash-max-fb)
- [MiMo-V2.6-Profailed26 Sep](https://evalmap.ai/jeep/#mimo-v26-pro-fb)
- [Solar Pro 4failed26 Sep](https://evalmap.ai/jeep/#solar-pro4-fb)
- [Solar Mini 4failed26 Sep](https://evalmap.ai/jeep/#solar-mini4-fb)
- [GLM-5.3-Flashfailed26 Sep](https://evalmap.ai/jeep/#glm53-flash-max-fb)
- [Gemma 4 31Bfailed26 Sep](https://evalmap.ai/jeep/#gemma4-31b-oc2)
- [Nemotron 3.5 Lightningfailed26 Sep](https://evalmap.ai/jeep/#nemotron35-lightning-or)
- [GPT-6 Lunamax effort26 Sep](https://evalmap.ai/jeep/#gpt6luna-max-fb)
- [MiMo 2.6 Flash26 Sep](https://evalmap.ai/jeep/#mimo-v26-flash-fb)
- [DeepSeek V4.1 Flashmax effort26 Sep](https://evalmap.ai/jeep/#deepseek-v41-flash-max-fb)
- [LongCat-2.5-Previewon the bench26 Sep](https://evalmap.ai/jeep/#longcat25-preview-oc2)
- [GLM-5.1high effort26 Sep](https://evalmap.ai/jeep/#glm51-high-oc2)
- [Qwen3 Coder Nexton the bench26 Sep](https://evalmap.ai/jeep/#qwen3-coder-next-oc2)
- [MiMo-V2.5high effort26 Sep](https://evalmap.ai/jeep/#mimo-v25-high-oc2)
- [MiniMax-M2.7on the bench26 Sep](https://evalmap.ai/jeep/#minimax-m27-oc2)
- [Gemini 3.6 Flashhigh effort26 Sep](https://evalmap.ai/jeep/#gemini36-flash-high-cmd)
- [Gemini 3.5 Flashfailed25 Sep](https://evalmap.ai/jeep/#gemini35-flash-high-cmd)
- [Inkling Smallfailed25 Sep](https://evalmap.ai/jeep/#inkling-small-cmd)
- [Kimi K2.7 Codefailed25 Sep](https://evalmap.ai/jeep/#kimi-k27-code-oc2)
- [MiMo-V2.5-Profailed25 Sep](https://evalmap.ai/jeep/#mimo-v25-pro-oc2)
- [Hy3failed25 Sep](https://evalmap.ai/jeep/#hy3-oc2)
- [Kimi K2.6failed25 Sep](https://evalmap.ai/jeep/#kimi-k26-oc2)
- [Ling-3.0-flashfailed25 Sep](https://evalmap.ai/jeep/#ling3-flash-oc2)
- [DeepSeek V4 Flash 0731failed25 Sep](https://evalmap.ai/jeep/#deepseek-v4-flash-0731-oc2)
- [Qwen3.5 397B A17Bfailed25 Sep](https://evalmap.ai/jeep/#qwen35-397b-oc2)
- [GLM-5.2failed25 Sep](https://evalmap.ai/jeep/#glm52-oc2)
- [MiniMax-M3failed25 Sep](https://evalmap.ai/jeep/#minimax-m3-oc2)
- [DeepSeek V4 Pro 0813failed25 Sep](https://evalmap.ai/jeep/#deepseek-v4-pro-0813-oc2)
- [GLM-5.3-Primefailed25 Sep](https://evalmap.ai/jeep/#glm53-prime-max-oc2)
- [Qwen3.7 Pluson the bench25 Sep](https://evalmap.ai/jeep/#qwen37-plus-cmd)
- [GPT-5.6 Lunamax effort25 Sep](https://evalmap.ai/jeep/#gpt56-luna-max-cmd)
- [GPT-5.6 Terramax effort25 Sep](https://evalmap.ai/jeep/#gpt56-terra-max-cmd)
- [Gemini 3.8 Flashhigh effort25 Sep](https://evalmap.ai/jeep/#gemini38-flash-high-cmd)
- [Claude Opus 5max effort25 Sep](https://evalmap.ai/jeep/#claude-opus-5-max-cmd)
- [Gemini 3.6 Flashdefault effort25 Sep](https://evalmap.ai/jeep/#gemini36-flash-cmd)
- [Inklingon the bench25 Sep](https://evalmap.ai/jeep/#inkling-cmd)
- [Qwen3.7 Maxon the bench25 Sep](https://evalmap.ai/jeep/#qwen37-cmd)
- [GPT-5.6 Solmax effort25 Sep](https://evalmap.ai/jeep/#gpt56-sol-max-cmd)
- [Claude Sonnet 5max effort25 Sep](https://evalmap.ai/jeep/#claude-sonnet-5-max-cmd)
- [Kimi K3max effort25 Sep](https://evalmap.ai/jeep/#kimi-k3-max-cmd)
- [Qwen3.8 Max Primemax effort25 Sep](https://evalmap.ai/jeep/#qwen38-max-prime)
- [Ling-3.0-flash-Finfailed24 Sep](https://evalmap.ai/jeep/#ling3-flash-fin-oc2)
- [Granite 4.2 8Bfailed24 Sep](https://evalmap.ai/jeep/#granite42-8b-high-oc2)
- [Mercury 2.5failed24 Sep](https://evalmap.ai/jeep/#mercury25-high-oc2)
- [Solar Mini 4failed24 Sep](https://evalmap.ai/jeep/#solar-mini4-max-oc2)
- [Ling-3.0-flash-VLon the bench24 Sep](https://evalmap.ai/jeep/#ling3-flash-vl-oc2)
- [DeepSeek-V4-Flash-Vision-Expmax effort24 Sep](https://evalmap.ai/jeep/#deepseek-v4-flash-vision-exp-max-oc2)
- [Qwen3.8-Flash-Nexton the bench24 Sep](https://evalmap.ai/jeep/#qwen38-flash-oc2)
- [AliceAI-Foundation-80B-A3B-Basefailed23 Sep](https://evalmap.ai/jeep/#aliceai-foundation-80b)
- [Space Bunny Alphamax effort23 Sep](https://evalmap.ai/jeep/#space-bunny-alpha-max)
- [GPT-6 Lunamax effort22 Sep](https://evalmap.ai/jeep/#gpt6luna-max-cx)
- [GPT-6 Solhigh effort22 Sep](https://evalmap.ai/jeep/#gpt6sol-high-cx)
- [Claude Opus 5.5xhigh effort22 Sep](https://evalmap.ai/jeep/#opus55-xhigh-cc)
- [GLM-5.3max effort22 Sep](https://evalmap.ai/jeep/#glm53-max-oc2)
- [MiMo-V2.6-Pro22 Sep](https://evalmap.ai/jeep/#mimo-v26-pro)
- [Fugu Max21 Sep](https://evalmap.ai/jeep/#fugu-max)
- [GLM-5.3-FlashX21 Sep](https://evalmap.ai/jeep/#glm53-flashx)
- [Grok 4.7xhigh effort21 Sep](https://evalmap.ai/jeep/#grok47-xhigh)

16–19 Sep

- [Step 5 Preview19 Sep](https://evalmap.ai/jeep/#step-5-preview-stepfun)
- [Pareto 26.916 Sep](https://evalmap.ai/jeep/#union-alpha-092026)

8–11 Sep

- [Kimi K2.8 Preview11 Sep](https://evalmap.ai/jeep/#kimi-k28preview)
- [Devin SWE-210 Sep](https://evalmap.ai/jeep/#devin-swe2)
- [MiMo-X-Pro-Preview9 Sep](https://evalmap.ai/jeep/#mimo-x-pro)
- [DeepSeek V4.1 Flash8 Sep](https://evalmap.ai/jeep/#deepseek-v4-1-flash-beta)

1–5 Sep

- [Qwen3.8 27Bon Cerebras5 Sep](https://evalmap.ai/jeep/#cerebrast-qwen38-27b-1300toksec-medium-effort)
- [Claude Fable 5.1medium effort5 Sep](https://evalmap.ai/jeep/#fable51-medium-effort)
- [GPT-6 Astra4 Sep](https://evalmap.ai/jeep/#astra)
- [Muse Spark 1.3max effort4 Sep](https://evalmap.ai/jeep/#muse13-max-effort)
- [Omen Alpha4 Sep](https://evalmap.ai/jeep/#omen-alpha-maybe-glm5)
- [Muse Spark 1.32 Sep](https://evalmap.ai/jeep/#muse13)
- [Gemini 3.8 Flashsecond build2 Sep](https://evalmap.ai/jeep/#aaa-jeep-game-gemini38flash)
- [Gemini 3.8 Flash2 Sep](https://evalmap.ai/jeep/#gemini38flash)
- [Claude Fable 5.12 Sep](https://evalmap.ai/jeep/#fable51)
- [Qwen3.8 Max 09022 Sep](https://evalmap.ai/jeep/#qwen38-max-0902)
- [Muse Spark 1.21 Sep](https://evalmap.ai/jeep/#muse12)
- [Qwen3.8 Flashon two DGX Sparks1 Sep](https://evalmap.ai/jeep/#qwen38flash-local-2xsparks)

26–28 Aug

- [HY4 Preview28 Aug](https://evalmap.ai/jeep/#tecent-hy4-preview)
- [GLM-5.3-Flashon two DGX Sparks27 Aug](https://evalmap.ai/jeep/#glm53-flash-2x-sparks-localai)
- [Qwen3.8 27Bon an RTX 409027 Aug](https://evalmap.ai/jeep/#qwen38-27b-local-4090rtx)
- [GLM-5.3-Flash26 Aug](https://evalmap.ai/jeep/#glm5-3-flash)
- [GLM-5.326 Aug](https://evalmap.ai/jeep/#glm53)
- [Gemini 3.7 Flash26 Aug](https://evalmap.ai/jeep/#gemini-3-7-flash)

18 Aug

- [Jeep Racemodel unknown18 Aug](https://evalmap.ai/jeep/#jeep-race-webgpu)

14 Aug

- [MudTrail 4x4model unknown14 Aug](https://evalmap.ai/jeep/#jeep-webgpu-race)

3 Aug

- [Jeep Racemodel unknown3 Aug](https://evalmap.ai/jeep/#jeep-race)

23–24 Jul

- [DUST RAIDERSmodel unknown24 Jul](https://evalmap.ai/jeep/#dust-raiders)
- [DESERT RAIDmodel unknown23 Jul](https://evalmap.ai/jeep/#desert-jeep-rally)

16–19 Jul

- [Jeep Road Runnermodel unknown19 Jul](https://evalmap.ai/jeep/#jeep-road-runner)
- [Jeep Rallymodel unknown19 Jul](https://evalmap.ai/jeep/#jeep-rally)
- [Desert Rallymodel unknown16 Jul](https://evalmap.ai/jeep/#jeep-racing)
- [Jeep Racingmodel unknown16 Jul](https://evalmap.ai/jeep/#jeep-rally-webgpu)

Newest first. Each card opens the build itself, served from this site; a build that failed the brief has nothing to open.

3–7 Nov 2026

JEEP RUSH — Canyon Rally

[A frame from the jeep race written by Qoduck (Qwen4 Flash/Plus/Max?) (xhigh)Play it ▶](https://evalmap.ai/jeep/qoduck-xhigh/)

Race frame at the start (0.05 s in, “GO!”) from a local software-WebGPU (SwiftShader) capture in headless Chrome, 7 November 2026; not a native RTX 5090 capture.

[Play it here](https://evalmap.ai/jeep/qoduck-xhigh/)not on the atlas

Jeep Racing — named in Russian by its author; this is the English rendering

[A frame from the jeep race written by Mercury 3Play it ▶](https://evalmap.ai/jeep/mercury3-high/)

Race frame at the start (0.5 s in) from a local software-WebGPU (SwiftShader) capture in headless Chrome, 6 November 2026; not a native RTX 5090 capture.

[Play it here](https://evalmap.ai/jeep/mercury3-high/)not on the atlas

Jeep Rush — WebGPU

[A frame from the jeep race written by Devin SWE-2 Flash (max)Play it ▶](https://evalmap.ai/jeep/devin-swe2-flash-max/)

Countdown frame from a local software-WebGPU (SwiftShader) capture in headless Chrome, 6 November 2026; not a native RTX 5090 capture.

[Play it here](https://evalmap.ai/jeep/devin-swe2-flash-max/)not on the atlas

game31 — Desert Canyon Circuit

[A frame from the jeep race written by LongCat 2.5 (max)Play it ▶](https://evalmap.ai/jeep/longcat25-max/)

Race frame from a local software-WebGPU (SwiftShader) capture in headless Chrome, 3 November 2026; not a native RTX 5090 capture.

[Play it here](https://evalmap.ai/jeep/longcat25-max/)not on the atlas

27 Oct

Jeep — Race — named in Russian by its author; this is the English rendering

[A frame from the jeep race written by GPT-6.1 Terra (max)Play it ▶](https://evalmap.ai/jeep/gpt61terra-max/)

Race frame from a local software-WebGPU (SwiftShader) capture in headless Chrome, 27 October 2026; not a native RTX 5090 capture.

[Play it here](https://evalmap.ai/jeep/gpt61terra-max/)not on the atlas

19–23 Oct

Ridge Runner — WebGPU Off-road

[A frame from the jeep race written by Kimi K3.1 Pro (max)Play it ▶](https://evalmap.ai/jeep/kimi-k31-pro-max/)

Countdown frame from a local software-WebGPU (SwiftShader) capture in headless Chrome, 23 October 2026; not a native RTX 5090 capture.

[Play it here](https://evalmap.ai/jeep/kimi-k31-pro-max/)not on the atlas

DUST RALLY — Jeep Off-Road Race

[A frame from the jeep race written by Kimi K2.7 CodePlay it ▶](https://evalmap.ai/jeep/kimi-k27-code/)

Countdown frame from a local WebGL fallback capture (headless Chromium), 19 October 2026; not a native RTX 5090 capture.

[Play it here](https://evalmap.ai/jeep/kimi-k27-code/)not on the atlas

Jeep Race — WebGPU

[A frame from the jeep race written by MiniMax-M3.1-FlashPlay it ▶](https://evalmap.ai/jeep/minimax-m31-flash/)

Countdown frame from a local WebGL 2 fallback capture (headless Chromium), 19 October 2026; not a native RTX 5090 capture.

[Play it here](https://evalmap.ai/jeep/minimax-m31-flash/)not on the atlas

12–18 Oct

Jeep Race — WebGPU

[A frame from the jeep race written by Beam 1.0 (max)Play it ▶](https://evalmap.ai/jeep/reflection-beam-max/)

Race frame from a local software-rendered (SwiftShader) capture, 18 October 2026; not a native RTX 5090 capture.

[Play it here](https://evalmap.ai/jeep/reflection-beam-max/)[this model on the atlas →](https://evalmap.ai/#reflection-beam)

Jeep Expedition — Game 14

[A frame from the jeep race written by DeepSeek V4.1 Flash 1017Play it ▶](https://evalmap.ai/jeep/deepseek-v41-flash-1017/)

Race frame from a local software-WebGL 2 (SwiftShader) capture, 18 October 2026; not a native RTX 5090 capture.

12/100

Rubric v4 · on this ladder that is less than a rudimentary demo, where 100 is a game on the level of GTA VI

- runs6/10
- world11/25
- driving8/20
- race7/15
- look8/15
- presentation8/15

Held to 12: in the test drives the jeep fell out of the world, drove off its edge, came apart or never moved, so the parts above, which add up to 48, do not count in full.

Measured at 58.6 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 2.2 with the processor slowed six times, and it drew with WebGPU.

An alpine valley of pines, rocks and a dirt loop under a snowy mountain ring, with three rivals, laps, minimap and sound. Driven straight on, the jeep climbs the ring and drives on past the ground into fog. WebGPU runs at 58.6 fps, 2.2 with CPU ×6.

[Play it here](https://evalmap.ai/jeep/deepseek-v41-flash-1017/)[this model on the atlas →](https://evalmap.ai/#deepseek-v4-1-flash-1017)

Jeep Race — Three.js WebGPU

A red jeep struck through: Nemotron 3.5 Super VL failed the brief — Dark screen: crash on first frame

Task failed. Nothing to drive. In a browser check over HTTP on 15 Oct, the WebGPU renderer it asks for, three@0.171.0/examples/jsm/renderers/webgpu/WebGPURenderer.js, is not in that release (the CDN answers 404), so the page falls back to WebGL as it says; then the first frame throws “Cannot read properties of undefined (reading 'setHex')”, because the checkpoint rings use MeshBasicMaterial, which has no emissive colour, and the loop stops. Only the HUD shows, over a dark screen, at speed 0 and 0:00.00. Published as the owner supplied it, unfixed; no playable copy.

Nothing to play[this model on the atlas →](https://evalmap.ai/#nemotron-3-5-super-vl)

CANYON OVERLAND RALLY 4x4

[A frame from the jeep race written by Gemini 4 Argon (max, cli2027)Play it ▶](https://evalmap.ai/jeep/gemini4-argon-max-cli2027/)

Race frame from a local software-WebGL 2 (SwiftShader) capture, 15 October 2026; not a native RTX 5090 capture.

[Play it here](https://evalmap.ai/jeep/gemini4-argon-max-cli2027/)[this model on the atlas →](https://evalmap.ai/#gemini-4-argon)

Alpine Canyon Jeep Rally

[A frame from the jeep race written by Gemini 4 Argon (max)Play it ▶](https://evalmap.ai/jeep/gemini4-argon-max/)

Race frame from a local software-WebGL 2 (SwiftShader) capture, 14 October 2026; not a native RTX 5090 capture.

[Play it here](https://evalmap.ai/jeep/gemini4-argon-max/)[this model on the atlas →](https://evalmap.ai/#gemini-4-argon)

Jeep Badlands Rally — named in Russian by its author; this is the English rendering

[A frame from the jeep race written by Devin SWE-2 (max)Play it ▶](https://evalmap.ai/jeep/devin-swe2-max/)

Race frame, native WebGPU on an RTX 5090 at 60 fps, 13 October 2026.

[Play it here](https://evalmap.ai/jeep/devin-swe2-max/)not on the atlas

Jeep Trailblazer — named in Russian by its author; this is the English rendering

[A frame from the jeep race written by Gemini 4 Argon (high)Play it ▶](https://evalmap.ai/jeep/gemini4-argon-high/)

Race frame, native WebGPU on an RTX 5090 at 60 fps, 13 October 2026.

[Play it here](https://evalmap.ai/jeep/gemini4-argon-high/)[this model on the atlas →](https://evalmap.ai/#gemini-4-argon)

Dust Rally

[A frame from the jeep race written by Muse Wonder Full (max)Play it ▶](https://evalmap.ai/jeep/muse-wonder-full-max/)

Countdown frame from a local software-WebGL (SwiftShader) capture, 12 October 2026; not a native RTX 5090 capture.

[Play it here](https://evalmap.ai/jeep/muse-wonder-full-max/)not on the atlas

5–8 Oct

Jeep Race — WebGPU

[A frame from the jeep race written by GLM-5.3-PrimePlay it ▶](https://evalmap.ai/jeep/glm53-prime-g22/)

Game frame from an eq2 software-WebGL pilot (SwiftShader, software WebGPU offered), 8 October 2026; not a native RTX 5090 capture.

Not scored: the standard RTX 5090 normal-speed and CPU ×6 drives were not run. Eq2 software-browser checks are not native-GPU performance measurements. The supplied source has no automated assertion test suite.

[Play it here](https://evalmap.ai/jeep/glm53-prime-g22/)not on the atlas

Jeep Race — WebGPU

[A frame from the jeep race written by Step 5 Preview (high)Play it ▶](https://evalmap.ai/jeep/step5-preview-high/)

Measured-drive frame on the RTX 5090 (WebGPU, 1280×720), 8 October 2026.

28/100

Rubric v4 · on this ladder that is a rudimentary demo, where 100 is a game on the level of GTA VI

- runs8/10
- world6/25
- driving4/20
- race3/15
- look2/15
- presentation5/15

Measured at 59.9 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 32.9 with the processor slowed six times, and it drew with WebGPU.

Hills, pines and rocks on rainbow ground: the shader passes its noise seed as mx_fractal_noise_vec3’s amplitude, so each colour channel swings far outside 0–1. A steep wall ends the world; no drive passed checkpoint 1; no rivals, no sound.

[Play it here](https://evalmap.ai/jeep/step5-preview-high/)[this model on the atlas →](https://evalmap.ai/#step-5)

Highland Rally

[A frame from the jeep race written by Claude Haiku 5.5Play it ▶](https://evalmap.ai/jeep/haiku55/)

Game frame from an eq2 software-WebGL pilot, 7 October 2026; not a native RTX 5090 capture.

Not scored: the standard RTX 5090 normal-speed and CPU ×6 drives were not run. Eq2 software-browser checks are not native-GPU performance measurements. The supplied source has no automated assertion test suite.

[Play it here](https://evalmap.ai/jeep/haiku55/)not on the atlas

Dustline Rally

[A frame from the jeep race written by Claude Fable 5.1 (xhigh)Play it ▶](https://evalmap.ai/jeep/fable51-xhigh/)

Game frame from an eq2 software-WebGL pilot, 7 October 2026; not a native RTX 5090 capture.

Not scored: the standard RTX 5090 normal-speed and CPU ×6 drives were not run. Eq2 software-browser checks are not native-GPU performance measurements. The supplied source has no automated assertion test suite; its screenshot script is not counted as a passing test.

[Play it here](https://evalmap.ai/jeep/fable51-xhigh/)[this model on the atlas →](https://evalmap.ai/#claude-fable-5-1-xhigh)

Thunder Ridge Rally

[A frame from the jeep race written by Claude Mythos 5.1 (xhigh)Play it ▶](https://evalmap.ai/jeep/mythos51-xhigh/)

Game frame from an eq2 software-WebGL pilot, 7 October 2026; not a native RTX 5090 capture.

Not scored: the standard RTX 5090 normal-speed and CPU ×6 drives were not run. Eq2 software checks are not native-GPU measurements. Seven source unit tests passed; the four-case source browser matrix had two passes and two HTTP-404 console failures.

[Play it here](https://evalmap.ai/jeep/mythos51-xhigh/)not on the atlas

Red Rock Rally

[A frame from the jeep race written by Mistral Large 4 PreviewPlay it ▶](https://evalmap.ai/jeep/mistral-large4-preview/)

48/100

Rubric v4 · on this ladder that is a solid hobby browser game, where 100 is a game on the level of GTA VI

- runs10/10
- world11/25
- driving8/20
- race4/15
- look7/15
- presentation8/15

Measured at 59.9 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 59.1 with the processor slowed six times, and it drew with WebGPU.

Desert slopes, mesas, textured sand and a styled HUD/minimap; floating scenery and blocky forms. WebGPU 59.9 fps, 59.1 CPU ×6; keeps moving, edge probe stayed on terrain. No completed lap/rival racing observed. WebGL fallback passed; two eq2 software-WebGPU tests timed out.

[Play it here](https://evalmap.ai/jeep/mistral-large4-preview/)[this model on the atlas →](https://evalmap.ai/#mistral-large-4-preview)

Jeep Race Game - Three.js

A red jeep struck through: Kolibri-1 failed the brief — Does not start: module error

Task failed. Does not start. In a CPU-only browser check, the untouched page stops on “Failed to resolve module specifier "three"”. No canvas, terrain or jeep appears; only the static HUD remains. Generation had stopped at the 26-minute bench limit. No playable copy is published.

The run OpenCode 2.0.12 · selfhosted/Kolibri-1

- Time: 26 min stopped at the bench’s time limit

- Tokens: 2.09M 0 read from cache · 190k written

- Cost: none per token the weights ran on a rented GPU, paid by the hour

- Work: 152 steps 1 session · 276 tool calls, 19 failed

Claude’s note. Public Kolibri-1 weights at Hugging Face revision e52eb4627d11516b0c01de49210ab5a4e4061444, served on a rented H200 NVL. Strict smoke passed; game generation timed out after 26 minutes, leaving index.html and main.js. The untouched page fails to resolve three and creates no canvas in a CPU-only browser with WebGPU disabled. A separate import-map/module-alias diagnostic copy then fails because BufferAttribute needs a Typed Array. Neither copy is published; no score or gameplay repair was made. The zero cost field is the harness token-price figure, not the GPU rental bill.

Nothing to playnot on the atlas

28 Sep–3 Oct

Ridge Runner — Alpine Expedition

[A frame from the jeep race written by Pareto 26.10 PreviewPlay it ▶](https://evalmap.ai/jeep/pareto-2610-preview/)

58/100

Rubric v4 · on this ladder that is a solid hobby browser game, where 100 is a game on the level of GTA VI

- runs9/10
- world12/25
- driving10/20
- race8/15
- look9/15
- presentation10/15

Measured at 59.9 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 46.6 with the processor slowed six times, and it drew with WebGPU.

An alpine valley, lake, pines, rocks and dirt loop; manual steering, three moving rivals, laps, standings, minimap and sound. Water and outer-world recovery return the jeep to the trail; the edge probe stayed on terrain. WebGPU runs at 59.9 fps, 46.6 with CPU ×6.

[Play it here](https://evalmap.ai/jeep/pareto-2610-preview/)not on the atlas

Saltwind Rally

[A frame from the jeep race written by Grok 4.7Play it ▶](https://evalmap.ai/jeep/grok47-build-fast-xhigh/)

40/100

Rubric v4 · on this ladder that is a solid hobby browser game, where 100 is a game on the level of GTA VI

- runs10/10
- world9/25
- driving4/20
- race6/15
- look5/15
- presentation6/15

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 59.3 with the processor slowed six times, and it drew with WebGPU.

A gravel loop through mesas, trees and a lagoon; two moving rivals, laps, positions, minimap and engine sound, at 60 fps on WebGPU even slowed. Throttle alone steers the jeep along the course; rivals overlap without collisions, but it stays on the ground and resets before the world's edge.

[Play it here](https://evalmap.ai/jeep/grok47-build-fast-xhigh/)[this model on the atlas →](https://evalmap.ai/#grok-4-7)

Dust Circuit — WebGPU Jeep Race

A red jeep struck through: Grok 4.7 failed the brief — Stuck on “Loading modules…”

Task failed. Nothing to drive. The page stops at its title card, “Building world — Loading modules…”, and stays there: js/main.js does not parse, because line 808 writes -(…) ** 2, which JavaScript refuses without parentheses, so no terrain, jeep or race ever appears.

The run OpenCode 2.0.12 · openrouter/x-ai/grok-4.7#xhigh

- Time: 1 h 1 min

- Tokens: 2.88M 2.38M read from cache · 123k written

- Cost: $2.68 pay-per-token, as OpenCode priced it

- Work: 33 steps 1 session · 44 tool calls, 1 failed

Claude’s note. Grok 4.7 at xhigh through OpenRouter. It spent its first 25 minutes reading three.js r181's sources and examples, then wrote index.html and a 71 KB js/main.js on three.js from jsDelivr without opening the page. It then started python3 -m http.server in the foreground; the call never returned, and after 22 idle minutes the run was stopped, which leaves the same files the time limit would have. The page stops at Loading modules: main.js line 808 writes -(...) ** 2, a syntax error in JavaScript, so no world or jeep ever appears.

Nothing to play[this model on the atlas →](https://evalmap.ai/#grok-4-7)

Jeep Race — WebGPU

A red jeep struck through: Fledge Alpha Free failed the brief — Blank page: three.js file is a 404

Task failed. Nothing to drive. Its one index.html imports three/tsl from three@0.170.0/build/three.tsl.js, a file that release does not have (the CDN answers 404), so the module graph never loads and the page stays empty: no terrain, no jeep.

The run OpenCode 2.0.14 · opencode/fledge-alpha-free

- Time: 2 min

- Tokens: 69k 48k read from cache · 8k written

- Cost: $0 a free model

- Work: 6 steps 1 session · 5 tool calls, 1 failed

Claude’s note. Free preview model on OpenCode Zen. It wrote a single index.html in about two minutes and said it was done without opening it in a browser. The page imports three/tsl from three@0.170.0, where no such file exists (404), so the module fails to load and nothing is drawn. The first attempt died on a provider 503 before the model wrote anything useful and is kept as aborted.

Nothing to playnot on the atlas

Jeep Trail Rush — WebGPU

[A frame from the jeep race written by Ling-3.1-flashPlay it ▶](https://evalmap.ai/jeep/ling31-flash-high-oc2/)

37/100

Rubric v4 · on this ladder that is a rudimentary demo, where 100 is a game on the level of GTA VI

- runs10/10
- world9/25
- driving4/20
- race3/15
- look5/15
- presentation6/15

Measured at 60.1 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 60 with the processor slowed six times, and it drew with WebGPU.

A valley under a snowy ring of mountains with a lake, pines, rocks and a dirt loop, timed laps with a best lap and a minimap, 60 fps on WebGPU even slowed; no rivals or checkpoints, the jeep passes through trees at up to 220 km/h, and its dust is flat grey squares.

The run OpenCode 2.0.12 · nano-gpt/inclusionai/ling-3.1-flash#high

- Time: 1 h 5 min

- Tokens: 23.83M 23.51M read from cache · 221k written

- Cost: $0.41 flat-rate plan: OpenCode’s price for the tokens, not a charge

- Work: 257 steps 1 session · 266 tool calls, 1 failed

Claude’s note. It worked on its own for 65 minutes over 257 steps; one of 266 tool calls failed. The game was written by the 20th minute: index.html, style.css and a main.js of 1,200 lines on three.js r170 from jsDelivr. It then installed puppeteer-core in the box and spent most of the run in its headless Chromium, which has no graphics card, reading three.js's WebGPU build to learn why the WebGL2 fallback drew nothing until it did, and closed with a long scripted drive and a check of the minimap. On a card it takes the WebGPU path: a valley under a snowy ring of mountains, a lake, a dirt loop, pines and rocks, timed laps with a best lap, but no rivals, and the jeep drives through the trees.

[Play it here](https://evalmap.ai/jeep/ling31-flash-high-oc2/)not on the atlas

Dustline · Offroad Rally

[A frame from the jeep race written by GPT-6.1 SolPlay it ▶](https://evalmap.ai/jeep/gpt61sol-xhigh/)

45/100

Rubric v4 · on this ladder that is a solid hobby browser game, where 100 is a game on the level of GTA VI

- runs9/10
- world10/25
- driving4/20
- race7/15
- look7/15
- presentation8/15

Measured at 60.1 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 44.3 with the processor slowed six times, and it drew with WebGPU.

An alpine valley with a gravel loop, pine forest, rocks and a lake; three laps against two rivals through twelve checkpoints, a trail map, three cameras and a polished menu, at 60 fps on WebGPU; the jeep passes through trees and rocks, and the mountains hold it at the edge.

[Play it here](https://evalmap.ai/jeep/gpt61sol-xhigh/)[this model on the atlas →](https://evalmap.ai/#gpt-6-1-sol-xhigh)

Highlands Rally — Jeep Race

[A frame from the jeep race written by Claude Sonnet 5.5Play it ▶](https://evalmap.ai/jeep/sonnet55-xhigh/)

48/100

Rubric v4 · on this ladder that is a solid hobby browser game, where 100 is a game on the level of GTA VI

- runs6/10
- world12/25
- driving6/20
- race8/15
- look8/15
- presentation8/15

Measured at 59 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 8.8 with the processor slowed six times, and it drew with WebGPU.

A highland valley eight kilometres across, with snowy peaks, a lake, a river and dense forest, and a three-lap race against seven named rivals with checkpoints, standings, a minimap and sound, at 59 fps on WebGPU; slowed, it drops to 9, and off the road the jeep crawls through the trees.

[Play it here](https://evalmap.ai/jeep/sonnet55-xhigh/)[this model on the atlas →](https://evalmap.ai/#claude-sonnet-5-5-xhigh)

Badlands Rally — Stage 9

[A frame from the jeep race written by Pixel CanaryPlay it ▶](https://evalmap.ai/jeep/pixel-canary-xhigh/)

12/100

Rubric v4 · on this ladder that is less than a rudimentary demo, where 100 is a game on the level of GTA VI

- runs6/10
- world9/25
- driving4/20
- race5/15
- look4/15
- presentation5/15

Held to 12: in the test drives the jeep fell out of the world, drove off its edge, came apart or never moved, so the parts above, which add up to 33, do not count in full.

Measured at 59.9 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 35.1 with the processor slowed six times, and it drew with WebGPU.

Mesas, rock spires and a snowy rim around a gravel stage of 14 gates and three laps, at 60 fps on WebGPU, but drawn into a 300×150 canvas and stretched, so every frame is a blur; driven on, the jeep crosses the rim and keeps going through empty sky.

[Play it here](https://evalmap.ai/jeep/pixel-canary-xhigh/)not on the atlas

21–27 Sep

Jeep Race

A red jeep struck through: Solar Mini 4 failed the brief — Blank sky: a variable out of scope

Task failed. Nothing to drive: a plain sky-blue page. The track code declares width inside a loop and uses it again outside it for the rails, so building the scene throws “width is not defined” and no frame is ever drawn. It worked for 73 minutes and 174 steps, most of them terminal commands.

The run Freebuff 0.0.203 · upstage/solar-mini4

- Time: 1 h 13 min

- Tokens: not counted Freebuff reports only the context: 195k at most, 14.01M over its 174 steps

- Cost: 5 Freebucks Freebuff’s own currency, 5 an hour of session; the owner’s paid access, not dollars

- Work: 174 steps 1 session · 172 tool calls, 13 failed

Claude’s note. A rerun on the owner's request; the first build never loaded three.js. 73 minutes and 174 steps, most of them terminal commands. The track code uses a variable outside the loop that declares it, so building the scene throws and the page stays plain sky blue.

Nothing to playnot on the atlas

Jeep Rally — WebGPU

A red jeep struck through: DeepSeek V4 Pro 0813 failed the brief — A HUD over black: a bad colour

Task failed. Nothing to see: the start screen opens, and behind it the HUD sits over a black screen at 0 km/h with the clock at 0:00. Setting up the gate cones, it hands plain numbers (0xffd700) to setColorAt, which wants a colour object, so setup stops on “color.toArray is not a function” before the jeep or the frame loop is ever made. It was done in eight minutes.

The run OpenCode 2.0.12 · openrouter/deepseek/deepseek-v4-pro-0813#max

- Time: 8 min

- Tokens: 935k 761k read from cache · 83k written

- Cost: $0.50 pay-per-token, as OpenCode priced it

- Work: 10 steps 1 session · 19 tool calls

Claude’s note. A rerun from scratch through OpenRouter after NanoGPT's weekly limit cut the first one off, at max effort. Eight minutes and ten steps: jeep-race/index.html and main.js with three.js r180 from jsDelivr. It passes plain numbers to setColorAt for the gate cones, so setup throws before the jeep or the frame loop is made, and the HUD sits over a black screen.

Nothing to play[this model on the atlas →](https://evalmap.ai/#deepseek-v4-pro)

Jeep Race — three.js WebGPU

A red jeep struck through: Solar Mini 4 failed the brief — Stuck at START: points not vectors

Task failed. Nothing to drive: the menu comes up and stays. The track is laid along a curve given plain [x, y, z] arrays where three.js wants vectors, so setting up the game throws “u.distanceToSquared is not a function”, and START then stops on the track that was never made (“Cannot read properties of undefined (reading ‘samples’)”). Not one frame of the world is drawn. It worked for 23 minutes and 173 steps.

The run OpenCode 2.0.12 · openrouter/upstage/solar-mini4#max

- Time: 23 min

- Tokens: 21.83M 20.80M read from cache · 159k written

- Cost: $0.18 pay-per-token, as OpenCode priced it

- Work: 173 steps 1 session · 172 tool calls, 13 failed

Claude’s note. A rerun on OpenRouter at max effort, after the provider stalled the previous attempt. 23 minutes and 173 steps: a Vite project, built in the box image because its config names the box's own path. The track curve is given plain arrays instead of vectors, so setup throws and START stops on a track that was never made; nothing of the world is drawn.

Nothing to playnot on the atlas

Jeep Race Game - WebGPU

A red jeep struck through: Nemotron 3.5 Lightning failed the brief — Python comments in JavaScript

Task failed. Nothing to drive: a blank page. Its one index.html loads three.js r160 from unpkg, but its module script uses Python-style comments (“# Large hills”), a syntax error in JavaScript, so none of it runs. It was done in 90 seconds.

The run OpenCode 2.0.12 · openrouter/nvidia/nemotron-3.5-lightning

- Time: 2 min

- Tokens: 579k 541k read from cache · 10k written

- Cost: $0.03 pay-per-token, as OpenCode priced it

- Work: 33 steps 1 session · 32 tool calls

Claude’s note. A rerun on the owner's request. Ninety seconds: one index.html on three.js r160 from unpkg, whose module script carries Python-style comments, a syntax error in JavaScript, so none of it runs.

Nothing to play[this model on the atlas →](https://evalmap.ai/#nemotron-3-5-lightning)

Jeep Race 3D

A red jeep struck through: Ling-3.0-flash-Fin failed the brief — A script that stops mid-line

Task failed. Nothing to drive. In 27 minutes it spent 174,000 output tokens on one 12 KB index.html whose script ends mid-code, so the browser rejects it (“Unexpected end of input”); it also imports "three" by bare name with no import map. The first run wrote no files at all.

The run OpenCode 2.0.12 · openrouter/inclusionai/ling-3.0-flash-fin

- Time: 27 min

- Tokens: 1.58M 1.31M read from cache · 176k written

- Cost: $0.05 pay-per-token, as OpenCode priced it

- Work: 33 steps 1 session · 33 tool calls, 5 failed

Claude’s note. A rerun on the owner's request; the first run wrote no files. This time 27 minutes and 174,000 output tokens went into one 12 KB index.html whose script stops mid-code and imports three by bare name, so nothing runs.

Nothing to playnot on the atlas

Jeep Race — Off-Road

A red jeep struck through: Solar Pro 4 failed the brief — Vite’s template, left in place

Task failed. Nothing to drive: a black page under a HUD reading zero. It created Vite’s starter project and wrote its own index.html with a Russian HUD, but left the starter’s main.js, which writes into an #app element the page no longer has, so it throws at load. No game code was written at all.

The run Freebuff 0.0.203 · upstage/solar-pro4

- Time: 5 min

- Tokens: not counted Freebuff reports only the context: 15k at most, 117k over its 9 steps

- Cost: 10 Freebucks Freebuff’s own currency, 10 an hour of session; the owner’s paid access, not dollars

- Work: 8 steps 1 session · 8 tool calls, 1 failed

Claude’s note. A rerun on the owner's request. Under five minutes: it created Vite's starter project and wrote its own index.html with a HUD, but left the starter's main.js, which writes into an element the page no longer has, so it throws at load. No game code was written.

Nothing to playnot on the atlas

Jeep Race - Three.js + WebGPU

A red jeep struck through: Granite 4.2 8B failed the brief — Blank: a bare “three” import

Task failed. Nothing to drive: a blank page. Its main.js imports "three" by bare name, and the page has no import map and no bundler to resolve it, so the browser refuses the module and nothing runs. It installed three.js with npm but never served or opened the page.

The run OpenCode 2.0.12 · openrouter/ibm-granite/granite-4.2-8b#high

- Time: 24 min

- Tokens: 739k 625k read from cache · 94k written

- Cost: $0.03 pay-per-token, as OpenCode priced it

- Work: 34 steps 1 session · 33 tool calls, 1 failed

Claude’s note. A rerun on the owner's request. In 24 minutes it installed three.js with npm and wrote an index.html and a main.js that imports three by bare name, with no bundler and no import map, and never served or opened the page, so the browser refuses the module and nothing runs.

Nothing to play[this model on the atlas →](https://evalmap.ai/#granite-4-2-8b)

No page written

A red jeep struck through: Mercury 2.5 failed the brief — No files: a page in the chat

Task failed. Nothing to drive: it wrote no files. In its one step, twelve seconds long, it answered with a whole index.html in the chat and called no tool, so nothing reached the folder.

The run OpenCode 2.0.12 · openrouter/inception/mercury-2.5#high

- Time: 12 s

- Tokens: 10k 0 read from cache · 3k written

- Cost: $0.00 pay-per-token, as OpenCode priced it

- Work: 1 step 1 session · 0 tool calls

Claude’s note. A rerun on the owner's request. One step of twelve seconds: it answered with a whole index.html in the chat and called no tool, so nothing reached the folder, the same kind of answer as the first run's.

Nothing to playnot on the atlas

No page written

A red jeep struck through: Kimi K2.6 failed the brief — No files: the plan only, again

Task failed. Nothing to drive: it wrote no files, the same as the first time. In one minute and two steps it listed the empty folder, wrote out its plan for the game in the chat, and ended its turn.

The run OpenCode 2.0.12 · nano-gpt/moonshotai/kimi-k2.6:thinking

- Time: 1 min

- Tokens: 24k 0 read from cache · 10k written

- Cost: $0.03 flat-rate plan: OpenCode’s price for the tokens, not a charge

- Work: 2 steps 1 session · 1 tool call

Claude’s note. A rerun on the owner's request, in the box that now has a browser. It wrote nothing again: in one minute and two steps it listed the empty folder, set out its plan for the game in the chat and ended its turn.

Nothing to playnot on the atlas

Jeep Off-Road Race 4x4 (Three.js WebGPU)

A red jeep struck through: Gemini 3.8 Flash failed the brief — Dark page: a variable out of scope

Task failed. Nothing to drive: a dark, empty page. The terrain code declares attempts with let inside one method and assigns it in others, so building the world throws “attempts is not defined” as the page loads. In under four minutes it wrote a Vite project in TypeScript and built it, but, with a browser now in the box, still never opened it.

The run Freebuff 0.0.203 · google/gemini-3.8-flash#max

- Time: 3 min

- Tokens: not counted Freebuff reports only the context: 60k at most, 1.33M over its 37 steps

- Cost: 80 Freebucks Freebuff’s own currency, 80 an hour of session; the owner’s paid access, not dollars

- Work: 37 steps 1 session · 37 tool calls

Claude’s note. A rerun on the owner's request, in the box that now has a browser. Under four minutes: a Vite project in TypeScript, built but never opened. The terrain code declares a variable with let in one method and assigns it in others, so the page throws at load and stays dark.

Nothing to play[this model on the atlas →](https://evalmap.ai/#gemini-3-8-flash)

JEEP RUSH — WebGPU Off-Road Racing

A red jeep struck through: Space Bunny Alpha failed the brief — Stuck loading: a call to nothing

Task failed. Nothing to drive: the JEEP RUSH loading screen, its bar at 97% on “rolling out the grid…”, and no further. Setting up the race calls this.updateParticleCamera, but updateParticleCamera is a function it imports, not a method of the game, so the last loading step throws and the race never opens. Its own Vite build carries the same line. In 83 minutes it wrote 6,500 lines with simulation and scene tests that run in node, looked for a browser, found none, and never opened the page.

The run OpenCode 2.0.12 · opencode/space-bunny-free#max

- Time: 1 h 23 min

- Tokens: 95.06M 94.44M read from cache · 446k written

- Cost: $0 a free model

- Work: 461 steps 1 session · 482 tool calls, 16 failed

Claude’s note. Space Bunny Alpha again, on the owner's request, through OpenCode Zen, where it is free (Freebuff does not offer it). 83 minutes and 461 steps on its own: a Vite project of about 6,500 lines with a procedural 1 km landscape, a filleted circuit, AI rivals, audio and node tests of its physics and scene. It looked for a browser, found none and never opened the page. The loader stops at 97%: setting up the race calls this.updateParticleCamera, an imported function it never made a method, so nothing can be driven.

Nothing to playnot on the atlas

4x4 Off-Road Jeep Rally - WebGPU Racing

[A frame from the jeep race written by Gemini 3.8 FlashPlay it ▶](https://evalmap.ai/jeep/gemini38-flash-high-agy/)

12/100

Rubric v4 · on this ladder that is less than a rudimentary demo, where 100 is a game on the level of GTA VI

- runs6/10
- world7/25
- driving1/20
- race4/15
- look5/15
- presentation6/15

Held to 12: in the test drives the jeep fell out of the world, drove off its edge, came apart or never moved, so the parts above, which add up to 29, do not count in full.

Measured at 59.9 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 17.2 with the processor slowed six times, and it drew with WebGPU.

A detailed hill course with rocks, trees, a START banner, a rival, nitro, sound and a minimap, but its drag outweighs its engine and the jeep never passes 1 km/h: it cannot be driven.

The run Antigravity CLI 1.2.12 · agy/gemini-3.8-flash-high

- Time: 6 min

- Tokens: 2.56M 2.08M read from cache · 73k written

- Cost: $0.74 at google/gemini-3.8-flash list price; paid through a flat-rate plan, so not a charge

- Work: 45 steps 1 session · 45 tool calls

Claude’s note. Gemini 3.8 Flash again on the owner's request, now in Google's Antigravity CLI at gemini-3.8-flash-high, its top effort there. Six minutes and 45 steps: a Vite project with rivals, nitro, three cameras, sound and a minimap over detailed hills. But its drag takes 1.2% of the speed every frame against an engine of 420 over 1650 kg, so the jeep tops out near 1 km/h and never gets going.

[Play it here](https://evalmap.ai/jeep/gemini38-flash-high-agy/)[this model on the atlas →](https://evalmap.ai/#gemini-3-8-flash)

Jeep Race — Three.js WebGPU

[A frame from the jeep race written by Hy3Play it ▶](https://evalmap.ai/jeep/hy3-oc2-r2/)

32/100

Rubric v4 · on this ladder that is a rudimentary demo, where 100 is a game on the level of GTA VI

- runs10/10
- world6/25
- driving5/20
- race3/15
- look4/15
- presentation4/15

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 57 with the processor slowed six times, and it drew with WebGPU.

Smooth WebGPU at 60 fps even slowed, over green rolling hills with pines and rocks, eight gates and a compass on an endless timed loop; a box jeep, no road, no rivals, no sound.

The run OpenCode 2.0.12 · openrouter/tencent/hy3#high

- Time: 7 min

- Tokens: 339k 242k read from cache · 43k written

- Cost: $0.04 pay-per-token, as OpenCode priced it

- Work: 8 steps 1 session · 7 tool calls

Claude’s note. A rerun from scratch through OpenRouter after NanoGPT's weekly limit cut the first one off; high effort, as max failed the bench's smoke test. Seven minutes and eight steps: one index.html and src/main.js with three.js from jsDelivr. Green rolling hills with pines and rocks, eight gates round a loop with a compass to the next, laps timed without end; no road, rivals or sound.

[Play it here](https://evalmap.ai/jeep/hy3-oc2-r2/)not on the atlas

Jeep Rally — Canyon Loop · WebGPU

[A frame from the jeep race written by GLM-5.3-PrimePlay it ▶](https://evalmap.ai/jeep/glm53-prime-max-oc2-r2/)

12/100

Rubric v4 · on this ladder that is less than a rudimentary demo, where 100 is a game on the level of GTA VI

- runs6/10
- world10/25
- driving4/20
- race6/15
- look6/15
- presentation6/15

Held to 12: in the test drives the jeep fell out of the world, drove off its edge, came apart or never moved, so the parts above, which add up to 38, do not count in full.

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 48.2 with the processor slowed six times, and it drew with WebGPU.

A loop under snowy peaks with a lake and three rivals that race, smooth even on a slowed processor, but the jeep comes apart as it drives, its body lifting off its wheels, so it is held to 12.

The run OpenCode 2.0.12 · openrouter/z-ai/glm-5.3-prime#max

- Time: 34 min

- Tokens: 11.07M 10.73M read from cache · 154k written

- Cost: $7.87 pay-per-token, as OpenCode priced it

- Work: 85 steps 1 session · 86 tool calls, 1 failed

Claude’s note. A rerun on the owner's request; the first build drew only its HUD. 34 minutes and 85 steps: a static page with three.js vendored, a loop through meadows and dirt under snowy peaks, a lake, pines, three rivals that race, a minimap and sound, at 60 frames a second on WebGPU and 48 with the processor slowed. But the jeep comes apart as it drives, its body flying up while the wheels stay on the ground, and GO! never leaves the screen.

[Play it here](https://evalmap.ai/jeep/glm53-prime-max-oc2-r2/)not on the atlas

Desert Run — Jeep Rally (WebGPU)

[A frame from the jeep race written by Space Bunny AlphaPlay it ▶](https://evalmap.ai/jeep/space-bunny-alpha-max-zen-r2/)

45/100

Rubric v4 · on this ladder that is a solid hobby browser game, where 100 is a game on the level of GTA VI

- runs6/10
- world12/25
- driving6/20
- race6/15
- look8/15
- presentation7/15

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 15.5 with the processor slowed six times, and it drew with WebGPU.

The best-looking build of the week, a carved and banked road across a lush island at 60 fps on WebGPU, with a menu and free roam, held back by a race with no rivals and walls the jeep ends up outside.

The run OpenCode 2.0.12 · opencode/space-bunny-free#max

- Time: 2 h 41 min

- Tokens: 131.45M 130.40M read from cache · 389k written

- Cost: $0 a free model

- Work: 486 steps 1 session · 499 tool calls, 25 failed

Claude’s note. A rerun on the owner's request, through OpenCode Zen, where it is free; the first Zen run stalled on its loading screen. 161 minutes and 486 steps on its own: a Vite project of about 40 modules, TSL materials and post-processing, a 1.44 km island with a lake and a banked road carved into it, a title menu with a free-roam mode, and headless physics tests. With a browser in the box it rendered its own test pages and checked them. There are no rivals, and in the pilot's minute the jeep ends up outside the track wall and runs along it off the road.

[Play it here](https://evalmap.ai/jeep/space-bunny-alpha-max-zen-r2/)not on the atlas

Jeep Race — WebGPU Off-Road

[A frame from the jeep race written by GLM-5.3-FlashPlay it ▶](https://evalmap.ai/jeep/glm53-flash-max-fb-r2/)

12/100

Rubric v4 · on this ladder that is less than a rudimentary demo, where 100 is a game on the level of GTA VI

- runs6/10
- world12/25
- driving5/20
- race4/15
- look7/15
- presentation5/15

Held to 12: in the test drives the jeep fell out of the world, drove off its edge, came apart or never moved, so the parts above, which add up to 39, do not count in full.

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 30.3 with the processor slowed six times, and it drew with WebGPU.

A varied 2 km world of lake, hills and dunes at 60 fps on WebGPU, but held at full throttle the jeep drives off its edge into a pale void, so it is held to 12.

The run Freebuff 0.0.203 · z-ai/glm-5.3-flash#max

- Time: 43 min

- Tokens: not counted Freebuff reports only the context: 50k at most, 815k over its 28 steps

- Cost: 5 Freebucks Freebuff’s own currency, 5 an hour of session; the owner’s paid access, not dollars

- Work: 28 steps 1 session · 27 tool calls

Claude’s note. A rerun on the owner's request; the first build hid behind its own no-WebGPU panel. 43 minutes and 28 steps: a varied 2 km world with a lake crossed by a dirt causeway, green hills with pines and sand dunes, twelve gates, a minimap and sound, at 60 frames a second on WebGPU. Held at full throttle the jeep leaves the world at 54 seconds and drives on into a pale void.

[Play it here](https://evalmap.ai/jeep/glm53-flash-max-fb-r2/)[this model on the atlas →](https://evalmap.ai/#glm-5-3-flash)

JEEP RACE — WebGPU

[A frame from the jeep race written by MiMo-V2.6-ProPlay it ▶](https://evalmap.ai/jeep/mimo-v26-pro-fb-r2/)

42/100

Rubric v4 · on this ladder that is a solid hobby browser game, where 100 is a game on the level of GTA VI

- runs6/10
- world12/25
- driving6/20
- race7/15
- look7/15
- presentation4/15

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 8.7 with the processor slowed six times, and it drew with WebGPU.

A rich hilly world with a lake and three rivals that really race, at 60 fps on WebGPU, but heavy on a slower processor, with GO! stuck on screen, a position that never changes and no sound.

The run Freebuff 0.0.203 · mimo/mimo-v2.6-pro

- Time: 32 min

- Tokens: not counted Freebuff reports only the context: 96k at most, 2.25M over its 37 steps

- Cost: 30 Freebucks Freebuff’s own currency, 30 an hour of session; the owner’s paid access, not dollars

- Work: 37 steps 1 session · 43 tool calls

Claude’s note. A rerun on the owner's request; the first build stopped at load. 32 minutes and 37 steps, checked with node simulations of the track and the race rather than in a browser. Hills, a lake, pines and a dirt loop with three rivals that do race, at 60 frames a second on WebGPU but under 9 with the processor slowed. GO! stays on screen all race, the position reads 4th throughout, and there is no sound.

[Play it here](https://evalmap.ai/jeep/mimo-v26-pro-fb-r2/)[this model on the atlas →](https://evalmap.ai/#mimo-v2-6-pro)

Offroad Jeep WebGPU Race

[A frame from the jeep race written by Gemini 3.5 FlashPlay it ▶](https://evalmap.ai/jeep/gemini35-flash-high-cmd-r2/)

12/100

Rubric v4 · on this ladder that is less than a rudimentary demo, where 100 is a game on the level of GTA VI

- runs2/10
- world7/25
- driving3/20
- race4/15
- look4/15
- presentation6/15

Held to 12: in the test drives the jeep fell out of the world, drove off its edge, came apart or never moved, so the parts above, which add up to 26, do not count in full.

Measured at 60.1 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 32 with the processor slowed six times, and it drew with WebGL2.

Three terrains and a canyon track, drawn with WebGL on purpose against the brief, and the jeep tumbles within seconds and never recovers, so it is held to 12.

The run Command Code 1.66.0 · commandcode/google/gemini-3.5-flash#high

- Time: 7 min

- Tokens: 3.86M 3.49M read from cache · 50k written

- Cost: $1.46 at google/gemini-3.5-flash list price; paid through a flat-rate plan, so not a charge

- Work: 68 steps 1 session · 66 tool calls

Claude’s note. A rerun on the owner's request, at high effort, the most Command Code allows for it. Seven minutes and 68 steps: a menu of three terrains, a canyon track with eight rivals that never move, three laps of eight checkpoints. It chose three.js's WebGL renderer on purpose, against the brief's WebGPU, and never opened the page: the jeep tumbles within seconds and does not recover.

[Play it here](https://evalmap.ai/jeep/gemini35-flash-high-cmd-r2/)not on the atlas

Jeep Race — WebGPU rally

[A frame from the jeep race written by MiniMax-M3.1-Flash-PreviewPlay it ▶](https://evalmap.ai/jeep/minimax-m31-flash-preview-max-mcode/)

12/100

Rubric v4 · on this ladder that is less than a rudimentary demo, where 100 is a game on the level of GTA VI

- runs5/10
- world14/25
- driving2/20
- race4/15
- look9/15
- presentation7/15

Held to 12: in the test drives the jeep fell out of the world, drove off its edge, came apart or never moved, so the parts above, which add up to 41, do not count in full.

Measured at 60.1 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 59.9 with the processor slowed six times, and it drew with WebGPU.

A detailed WebGPU world with a real road, pines and rocks, a minimap, gauges and two rivals on the grid, but a TypeError from the dust effect stops the game the moment the jeep moves: it never leaves the start line, so it is held to 12.

The run MiniMax Code 0.5.5 · minimax/MiniMax-M3.1-Flash-Preview#max

- Time: 1 h 35 min

- Tokens: 87.00M 86.53M read from cache · 349k written

- Cost: $0 MiniMax Code priced every request at $0 on the owner’s account; no list price is published for this model, so it is not called free

- Work: 293 steps 1 session · 293 tool calls, 43 failed

Claude’s note. One MiniMax Code session at max effort, 95 minutes and 293 requests on the day the model opened there. It vendored three.js, wrote twenty modules and its own tests, drove an autopilot lap in Node and checked the page in a headless browser, then ended by itself. The world is the most detailed of the week: a 2.4 km map of eroded ridges, lakes and 1.7 million triangles, a 6 km track, pines, boulders and grass, at 60 frames a second on WebGPU. But the dust puffs assign a number to a sprite's read-only rotation, so the moment the jeep rolls off the line the update loop throws every frame: it freezes at 7 km/h with the clock at zero and never moves.

[Play it here](https://evalmap.ai/jeep/minimax-m31-flash-preview-max-mcode/)not on the atlas

Jeep Race (WebGPU + Three.js)

A red jeep struck through: gpt-oss-120b failed the brief — White page: an import 404s

Task failed. Nothing to drive: a blank white page. The game imports simplex-noise@4.0.0/simplex-noise.js from jsDelivr, a file that package does not have (404), so its module never runs. The same on the untouched page with its CDN files.

The run OpenCode 2.0.12 · nano-gpt/openai/gpt-oss-120b#high

- Time: 3 min

- Tokens: 24k 32 read from cache · 4k written

- Cost: $0.01 flat-rate plan: OpenCode’s price for the tokens, not a charge

- Work: 4 steps 1 session · 2 tool calls

Claude’s note. High reasoning effort, the highest listed. In 3 minutes over 4 steps it wrote an HTML page and one module; NanoGPT answered its last step with 'Upstream emitted malformed tool call data', after the model had written its closing message. The game does not start, also on its untouched CDN page.

Nothing to playnot on the atlas

4x4 Jeep Offroad Rally - Three.js WebGPU

A red jeep struck through: Gemini 3.8 Flash failed the brief — Only the HUD: the loop throws

Task failed. Nothing to drive: its HUD, in Russian, over a black page. The camera code calls subScaledVector on a three.js vector, a method three.js does not have, so the game loop throws on its first frame and on every one after, and nothing of the world is ever drawn. It built the project with Vite in under three minutes and did not open it.

The run Freebuff 0.0.203 · google/gemini-3.8-flash#max

- Time: 3 min

- Tokens: not counted Freebuff reports only the context: 47k at most, 639k over its 26 steps

- Cost: 80 Freebucks Freebuff’s own currency, 80 an hour of session; the owner’s paid access, not dollars

- Work: 26 steps 1 session · 26 tool calls

Claude’s note. One Freebuff session at max reasoning: under three minutes and 26 steps, a Vite project with an AI rival, a sound manager and a Russian HUD, built but never opened. Its camera follower calls a vector method three.js does not have, so the loop throws from the first frame and the page shows only the HUD over black.

Nothing to play[this model on the atlas →](https://evalmap.ai/#gemini-3-8-flash)

JEEP RACE — WebGPU

A red jeep struck through: MiMo-V2.6-Pro failed the brief — Stops at load: noise misused

Task failed. Nothing to drive: a dark page with one line, in Russian, “Could not start the game, e.random is not a function”. Its noise module hands three.js’s SimplexNoise a bare random function where that class takes an object with a random() method, so building the world stops at once and nothing is drawn. It built the project with Vite but never opened it in a browser.

The run Freebuff 0.0.203 · mimo/mimo-v2.6-pro

- Time: 35 min

- Tokens: not counted Freebuff reports only the context: 85k at most, 683k over its 15 steps

- Cost: 30 Freebucks Freebuff’s own currency, 30 an hour of session; the owner’s paid access, not dollars

- Work: 14 steps 1 session · 30 tool calls

Claude’s note. One Freebuff session, no reasoning setting offered: 35 minutes, 14 steps, a Vite project of twelve modules that it built without error but never opened. At load it passes a bare function to three.js's SimplexNoise, which wants an object with random(), and the page stops on that error before drawing anything.

Nothing to play[this model on the atlas →](https://evalmap.ai/#mimo-v2-6-pro)

Jeep Race — Three.js + WebGPU

A red jeep struck through: Solar Pro 4 failed the brief — Black behind its menu

Task failed. Nothing to drive: the “JEEP RACE” menu over a black page, and START ENGINE starts nothing. Building the world stops at load with “Cannot read properties of undefined (reading ‘x’)” in its catmullRom spline, called while the track road is laid, so there is no ground, track or jeep. It renders with WebGL; its code sets preferWebGPU = false.

The run Freebuff 0.0.203 · upstage/solar-pro4

- Time: 5 min

- Tokens: not counted Freebuff reports only the context: 32k at most, 176k over its 9 steps

- Cost: 10 Freebucks Freebuff’s own currency, 10 an hour of session; the owner’s paid access, not dollars

- Work: 9 steps 1 session · 9 tool calls

Claude’s note. One Freebuff session, no reasoning setting offered: five minutes, nine steps, four files written in one pass and never run. It chose WebGL itself, setting preferWebGPU to false, and the world build throws in its Catmull-Rom spline at load, so the menu sits over a black page.

Nothing to playnot on the atlas

Jeep Race — Three.js WebGPU

A red jeep struck through: Solar Mini 4 failed the brief — Menu only: three.js never loads

Task failed. Nothing to drive: a “Jeep Race” menu whose Start race button does nothing, and no canvas. Its scripts import three.js by the bare name “three”, with no import map and no bundler, so the browser refuses them (“Failed to resolve module specifier”) and no game code runs. Checked served from its own folder, as its npm run dev would serve it.

The run Freebuff 0.0.203 · upstage/solar-mini4

- Time: 13 min

- Tokens: not counted Freebuff reports only the context: 142k at most, 13.81M over its 136 steps

- Cost: 5 Freebucks Freebuff’s own currency, 5 an hour of session; the owner’s paid access, not dollars

- Work: 136 steps 1 session · 135 tool calls, 6 failed

Claude’s note. One Freebuff session, which offers no reasoning setting for Solar Mini 4: 13 minutes, 136 steps and 96 shell commands, mostly node --check and curl against its own server. It never loaded the page in a browser, where the bare import of three with no import map stops every script, leaving a menu whose Start race button does nothing.

Nothing to playnot on the atlas

Jeep Race — three.js WebGPU

A red jeep struck through: GLM-5.3-Flash failed the brief — Hidden by its NO WEBGPU panel

Task failed. Nothing to see: every frame is its own “NO WEBGPU — This game needs a WebGPU browser” panel, on a browser where WebGPU works. The panel carries the hidden attribute, but its style sheet gives it display: flex, which overrides that, and it sits over the whole page above the game. Underneath, the game runs on WebGPU at 60 frames a second and the jeep can be driven through its checkpoints, unseen.

The run Freebuff 0.0.203 · z-ai/glm-5.3-flash#max

- Time: 33 min

- Tokens: not counted Freebuff reports only the context: 46k at most, 672k over its 23 steps

- Cost: 5 Freebucks Freebuff’s own currency, 5 an hour of session; the owner’s paid access, not dollars

- Work: 23 steps 1 session · 24 tool calls

Claude’s note. One Freebuff session at max reasoning, 33 minutes and 23 steps, ended by itself after checking its files with node but never in a browser. The game underneath is real: WebGPU at 60 frames a second, a 2.4 km carved loop, checkpoints and three cameras. But its own NO WEBGPU panel stays over the whole page, because the style sheet's display: flex overrides the hidden attribute, so a player sees only that panel.

Nothing to play[this model on the atlas →](https://evalmap.ai/#glm-5-3-flash)

WebGPU Jeep Race

A red jeep struck through: Gemma 4 31B failed the brief — Black screen and a speed readout

Task failed. Nothing to drive: a black page with “Speed: 0 km/h”. Its setup calls renderer.setToneMapping, which three.js’s WebGPURenderer does not have, so it stops with “renderer.setToneMapping is not a function” before the first frame. The same on the untouched page with its CDN files.

The run OpenCode 2.0.12 · nano-gpt/google/gemma-4-31b-it:thinking

- Time: 4 min

- Tokens: 55k 23k read from cache · 10k written

- Cost: $0.01 flat-rate plan: OpenCode’s price for the tokens, not a charge

- Work: 4 steps 1 session · 7 tool calls, 2 failed

Claude’s note. Reasoning on (the :thinking route); NanoGPT lists no effort levels for it. In 4 minutes over 4 steps it wrote an HTML page, a script and a style sheet; the game does not start, also on its untouched CDN page.

Nothing to playnot on the atlas

Jeep Race Game - WebGPU

A red jeep struck through: Nemotron 3.5 Lightning failed the brief — Frozen: the loop fails each frame

Task failed. Nothing to drive. The hills and the course draw once, then the game loop stops on every frame with “flWheel is not defined”: the wheels are made inside the function that builds the jeep and used outside it, so nothing moves and the clock stays at 0:00. The same on the untouched page with its CDN files.

The run OpenCode 2.0.12 · openrouter/nvidia/nemotron-3.5-lightning

- Time: 2 min

- Tokens: 612k 496k read from cache · 24k written

- Cost: $0.03 pay-per-token, as OpenCode priced it

- Work: 17 steps 1 session · 16 tool calls, 4 failed

Claude’s note. Run through OpenRouter: on NanoGPT, the subscription route, the model timed out twice before it wrote anything (25 and 26 Sep; kept aside, not runs). On OpenRouter its reasoning is only on or off, so no effort level was set. In 2 minutes over 17 steps it wrote one HTML page; the game does not start, also on its untouched CDN page.

Nothing to play[this model on the atlas →](https://evalmap.ai/#nemotron-3-5-lightning)

Ridgeline Rally — Wild Country

[A frame from the jeep race written by GPT-6 LunaPlay it ▶](https://evalmap.ai/jeep/gpt6luna-max-fb/)

12/100

Rubric v4 · on this ladder that is less than a rudimentary demo, where 100 is a game on the level of GTA VI

- runs6/10
- world8/25
- driving5/20
- race4/15
- look7/15
- presentation5/15

Held to 12: in the test drives the jeep fell out of the world, drove off its edge, came apart or never moved, so the parts above, which add up to 35, do not count in full.

Measured at 59.9 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 22.7 with the processor slowed six times, and it drew with WebGPU.

The most polished interface of its batch over a small low-poly loop, but a 1 km world with no edge: held at full throttle the jeep drives off it into a black void, so it is held to 12.

The run Freebuff 0.0.203 · openai/gpt-6-luna#max

- Time: 10 min

- Tokens: not counted Freebuff reports only the context: 74k at most, 942k over its 19 steps

- Cost: 20 Freebucks Freebuff’s own currency, 20 an hour of session; the owner’s paid access, not dollars

- Work: 19 steps 1 session · 26 tool calls

Claude’s note. One Freebuff session at max reasoning: ten minutes and 19 steps. It never installed its dependencies, so its own npm run build failed, and it said so and checked the syntax with node instead. Vite's default target then rejects its top-level await; the copy here is built with --target esnext, the code untouched. It has the most polished interface of the batch and a low-poly gravel loop, but the world is 1 km across with no edge: hold the throttle and the jeep drives off it into a black void.

[Play it here](https://evalmap.ai/jeep/gpt6luna-max-fb/)[this model on the atlas →](https://evalmap.ai/#gpt-6-luna)

Jeep Rally 4x4 — three.js WebGPU

[A frame from the jeep race written by MiMo 2.6 FlashPlay it ▶](https://evalmap.ai/jeep/mimo-v26-flash-fb/)

46/100

Rubric v4 · on this ladder that is a solid hobby browser game, where 100 is a game on the level of GTA VI

- runs8/10
- world11/25
- driving7/20
- race8/15
- look7/15
- presentation5/15

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 41.4 with the processor slowed six times, and it drew with WebGPU.

A real race on WebGPU at 60 fps: three AI rivals, three laps of twelve checkpoints, snowy hills and pines, but silent, a bare plain past the course, and a jeep that wedges on slopes.

The run Freebuff 0.0.203 · mimo/mimo-v2.5

- Time: 1 h 15 min stopped at the bench’s time limit

- Tokens: not counted Freebuff reports only the context: 231k at most, 14.66M over its 111 steps

- Cost: 10 Freebucks Freebuff’s own currency, 10 an hour of session; the owner’s paid access, not dollars

- Work: 110 steps 1 session · 124 tool calls, 1 failed

Claude’s note. One Freebuff session; Freebuff offers no reasoning setting for it and sends it as mimo/mimo-v2.5. The bench stopped it at 75 minutes, a limit set for Freebuff's hour-long session before we saw that Freebuff lets a working agent finish: it was still running its own browser tests of the gameplay, 110 steps in. Published as it stood. It is a real race: WebGPU at 60 frames a second, three AI rivals over three laps of twelve checkpoints, a start gate, minimap and pause, but no sound; the jeep can wedge itself on snowy slopes.

[Play it here](https://evalmap.ai/jeep/mimo-v26-flash-fb/)not on the atlas

Jeep Rally — three.js WebGPU

[A frame from the jeep race written by DeepSeek V4.1 FlashPlay it ▶](https://evalmap.ai/jeep/deepseek-v41-flash-max-fb/)

40/100

Rubric v4 · on this ladder that is a solid hobby browser game, where 100 is a game on the level of GTA VI

- runs7/10
- world11/25
- driving7/20
- race5/15
- look5/15
- presentation5/15

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 25.2 with the processor slowed six times, and it drew with WebGPU.

A varied 3.2 km world of ridges, river and lake at 60 fps, heavy on a slow processor; a time trial with no rivals or sound, broken strips of geometry in the sky, and a jeep that drives on water.

The run Freebuff 0.0.203 · deepseek/deepseek-v4-flash#max

- Time: 1 h 20 min

- Tokens: not counted Freebuff reports only the context: 258k at most, 16.00M over its 105 steps

- Cost: 15 Freebucks Freebuff’s own currency, 15 an hour of session; the owner’s paid access, not dollars

- Work: 105 steps 1 session · 138 tool calls, 1 failed

Claude’s note. One Freebuff session at max reasoning; it worked 80 minutes, on past the session's hour, wrote fifteen modules and its own driving and scene tests, and ended by itself. The world is 3.2 km of snowy ridges, a river, a lake and pines at 60 frames a second on WebGPU, with a countdown, three laps of eight checkpoints and a good HUD, but no rivals and no sound. Long stretched strips of geometry hang in the sky, debris floats over the lake, and the jeep drives straight across the water.

[Play it here](https://evalmap.ai/jeep/deepseek-v41-flash-max-fb/)not on the atlas

Jeep WebGPU Race

[A frame from the jeep race written by LongCat-2.5-PreviewPlay it ▶](https://evalmap.ai/jeep/longcat25-preview-oc2/)

32/100

Rubric v4 · on this ladder that is a rudimentary demo, where 100 is a game on the level of GTA VI

- runs9/10
- world7/25
- driving5/20
- race3/15
- look4/15
- presentation4/15

Measured at 60.1 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 47.5 with the processor slowed six times, and it drew with WebGPU.

Plays on WebGPU at 60 fps with detailed rolling terrain, but there is no road, no rivals or finish, no sound, and flat untextured looks under a stuck debug line.

The run OpenCode 2.0.12 · opencode/longcat-2.5-preview-free

- Time: 3 min

- Tokens: 146k 113k read from cache · 11k written

- Cost: $0 a free model

- Work: 9 steps 1 session · 13 tool calls, 1 failed

Claude’s note. Three minutes and nine steps, free on OpenCode Zen: one index.html written in a single go, then a few edits and a syntax check of its script; it never opened the page. It draws with WebGPU at 60 frames a second: rolling procedural hills coloured by height and slope, a lake, instanced cone pines, bushes and rocks, fog. The jeep drives, climbs and takes air, but there is no road, only twelve checkpoint markers that a lap silently resets; no rivals, no sound, and a debug line reading WebGPU renderer ready stays in the middle of the screen.

[Play it here](https://evalmap.ai/jeep/longcat25-preview-oc2/)not on the atlas

Jeep Rally — WebGPU

[A frame from the jeep race written by GLM-5.1Play it ▶](https://evalmap.ai/jeep/glm51-high-oc2/)

12/100

Rubric v4 · on this ladder that is less than a rudimentary demo, where 100 is a game on the level of GTA VI

- runs6/10
- world7/25
- driving3/20
- race5/15
- look3/15
- presentation5/15

Held to 12: in the test drives the jeep fell out of the world, drove off its edge, came apart or never moved, so the parts above, which add up to 29, do not count in full.

Measured at 60.1 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 39 with the processor slowed six times, and it drew with WebGPU.

A lake below snowy peaks with pines, laps, checkpoints and a route minimap, but shadow acne stripes all the ground; driven straight on, the jeep runs past the terrain onto a white fog plane at 20 seconds.

The run OpenCode 2.0.12 · nano-gpt/z-ai/glm-5.1:thinking#high

- Time: 46 min

- Tokens: 11.06M 10.06M read from cache · 91k written

- Cost: $2.42 flat-rate plan: OpenCode’s price for the tokens, not a charge

- Work: 155 steps 1 session · 153 tool calls, 2 failed

Claude’s note. High reasoning effort, the highest NanoGPT lists for it. The first attempt hit NanoGPT's 296-second request limit before the model wrote anything, so it is not a run and was run again. In 46 minutes over 155 steps it wrote an index.html and eleven modules under src/, with three.js r169's WebGPU build from jsDelivr through an import map; the page draws with WebGPU. Vendored for the page.

[Play it here](https://evalmap.ai/jeep/glm51-high-oc2/)not on the atlas

Jeep Race Game - Three.js WebGPU

[A frame from the jeep race written by Qwen3 Coder NextPlay it ▶](https://evalmap.ai/jeep/qwen3-coder-next-oc2/)

12/100

Rubric v4 · on this ladder that is less than a rudimentary demo, where 100 is a game on the level of GTA VI

- runs6/10
- world5/25
- driving1/20
- race3/15
- look3/15
- presentation4/15

Held to 12: in the test drives the jeep fell out of the world, drove off its edge, came apart or never moved, so the parts above, which add up to 22, do not count in full.

Measured at 60.1 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 33.1 with the processor slowed six times, and it drew with WebGL2.

Rocks, pines and patches of snow on brown dirt, a start menu, three laps and touch buttons, drawn with WebGL2; in both drives the jeep lifts off at 61 km/h and flies on through empty sky from about 20 seconds.

The run OpenCode 2.0.12 · nano-gpt/qwen/qwen3-coder-next

- Time: 46 s

- Tokens: 36k 0 read from cache · 8k written

- Cost: $0.02 flat-rate plan: OpenCode’s price for the tokens, not a charge

- Work: 3 steps 1 session · 2 tool calls

Claude’s note. A model without reasoning. In 46 seconds over 3 steps it wrote one file, jeep-race-game.html, published here as the page's index.html and otherwise unchanged; three.js r160 from a CDN, vendored; the page draws with WebGL. Its start screen is in Russian.

[Play it here](https://evalmap.ai/jeep/qwen3-coder-next-oc2/)not on the atlas

Jeep Race - WebGPU

[A frame from the jeep race written by MiMo-V2.5Play it ▶](https://evalmap.ai/jeep/mimo-v25-high-oc2/)

12/100

Rubric v4 · on this ladder that is less than a rudimentary demo, where 100 is a game on the level of GTA VI

- runs4/10
- world7/25
- driving2/20
- race4/15
- look5/15
- presentation5/15

Held to 12: in the test drives the jeep fell out of the world, drove off its edge, came apart or never moved, so the parts above, which add up to 27, do not count in full.

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 23.4 with the processor slowed six times, and it drew with WebGL2.

A green island of trees, grass and rocks in the sea, a GO countdown, laps and a minimap; the jeep crawls at 20 km/h, rolls off the shore and drives on across open water for the rest of the minute.

The run OpenCode 2.0.12 · nano-gpt/xiaomi/mimo-v2.5:thinking#high

- Time: 9 min

- Tokens: 189k 162k read from cache · 19k written

- Cost: $0.01 flat-rate plan: OpenCode’s price for the tokens, not a charge

- Work: 11 steps 1 session · 8 tool calls, 1 failed

Claude’s note. High reasoning effort, the highest NanoGPT lists for it. In 9 minutes over 11 steps it wrote one HTML page; its WebGPU renderer is imported from a path three.js r170 does not have, so the page falls back to its own WebGL path. NanoGPT answered its last step with 'Upstream emitted malformed tool call data', after the model had written its closing message.

[Play it here](https://evalmap.ai/jeep/mimo-v25-high-oc2/)not on the atlas

Jeep Race 3D

[A frame from the jeep race written by MiniMax-M2.7Play it ▶](https://evalmap.ai/jeep/minimax-m27-oc2/)

22/100

Rubric v4 · on this ladder that is a rudimentary demo, where 100 is a game on the level of GTA VI

- runs7/10
- world5/25
- driving2/20
- race2/15
- look3/15
- presentation3/15

Measured at 60.1 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 52.6 with the processor slowed six times, and it drew with WebGL2.

Green and sand hills with rocks and a few pines under fog, drawn with WebGL2; the jeep climbs whatever it meets with the speedometer pinned at 288, the camera sinks into slopes, and a clock counts up with nothing to race.

The run OpenCode 2.0.12 · nano-gpt/minimax/minimax-m2.7

- Time: 3 min

- Tokens: 186k 147k read from cache · 12k written

- Cost: $0.05 flat-rate plan: OpenCode’s price for the tokens, not a charge

- Work: 12 steps 1 session · 13 tool calls

Claude’s note. NanoGPT lists no effort setting for it; run once at the default. In 3 minutes over 12 steps it wrote one HTML page, and a SPEC.md beside it, with three.js r160 from a CDN; the page draws with WebGL. Vendored for the page.

[Play it here](https://evalmap.ai/jeep/minimax-m27-oc2/)not on the atlas

Detailed Terrain Jeep Race (WebGPU / Three.js)

[A frame from the jeep race written by Gemini 3.6 FlashPlay it ▶](https://evalmap.ai/jeep/gemini36-flash-high-cmd/)

12/100

Rubric v4 · on this ladder that is less than a rudimentary demo, where 100 is a game on the level of GTA VI

- runs6/10
- world5/25
- driving2/20
- race5/15
- look4/15
- presentation5/15

Held to 12: in the test drives the jeep fell out of the world, drove off its edge, came apart or never moved, so the parts above, which add up to 27, do not count in full.

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 46.9 with the processor slowed six times, and it drew with WebGPU.

Sand and grass hills under distant peaks, a start gantry, three laps with a best time and a route minimap; the jeep bounces off a rise and is in empty sky within 20 seconds, speedometer frozen at 115.

The run Command Code 1.50.1 · commandcode/google/gemini-3.6-flash#high

- Time: 3 min

- Tokens: 1.03M 818k read from cache · 23k written

- Cost: $0.29 at google/gemini-3.6-flash list price; paid through a flat-rate plan, so not a charge

- Work: 33 steps 1 session · 32 tool calls, 2 failed

Claude’s note. High reasoning effort, the highest Command Code offers for it; run again at the owner's request after the first run went at the CLI's default. It worked for 3 minutes over 33 steps and wrote a Vite project, built with vite build for the page.

[Play it here](https://evalmap.ai/jeep/gemini36-flash-high-cmd/)not on the atlas

Jeep WebGPU Racer - Offroad Challenge

A red jeep struck through: Gemini 3.5 Flash failed the brief — START leads nowhere

Task failed. Nothing to drive. The menu opens, but the race never starts: the page loads three.js r169’s three.webgpu.js with a plain script tag, the browser refuses that ES module (“Unexpected token ‘export’”), and the game’s code then stops on “THREE is not defined”. The same on the untouched page with its CDN files.

The run Command Code 1.50.1 · commandcode/google/gemini-3.5-flash#high

- Time: 6 min

- Tokens: 939k 759k read from cache · 53k written

- Cost: $0.78 at google/gemini-3.5-flash list price; paid through a flat-rate plan, so not a charge

- Work: 19 steps 1 session · 18 tool calls

Claude’s note. High reasoning effort: Command Code refuses max for this model. The game does not start, also on its untouched CDN page.

Nothing to playnot on the atlas

Jeep Race — WebGPU

A red jeep struck through: Inkling Small failed the brief — Stays on its start screen

Task failed. Nothing to drive. The start screen comes up, and START ENGINE throws “THREE.WebGPURenderer is not a constructor”: the page loads three.js r160 from unpkg, a release with no WebGPU renderer, and has no other path, so no canvas is ever made. The same on the untouched page with its CDN files.

The run Command Code 1.50.1 · commandcode/thinkingmachines/inkling-small

- Time: 22 s

- Tokens: 59k 39k read from cache · 7k written

- Cost: 0.02 credits Command Code’s own unit, paid through a flat-rate plan; no list price is published for this model

- Work: 3 steps 1 session · 2 tool calls

Claude’s note. No effort setting for this model on Command Code; run once at the CLI's default. models.dev lists no price for it. The game does not start, also on its untouched CDN page.

Nothing to play[this model on the atlas →](https://evalmap.ai/#inkling-small)

No page written

A red jeep struck through: Kimi K2.7 Code failed the brief — No files: it gave up at once

Task failed. Nothing to drive. It wrote no files. A web search failed and unpkg answered with a server error; it then fetched the same file from jsDelivr, and replied “The tool returned an error, so I could not complete the request.” The run ended there, after four steps and 104 seconds.

The run OpenCode 2.0.12 · nano-gpt/moonshotai/kimi-k2.7-code

- Time: 2 min

- Tokens: 37k 9k read from cache · 2k written

- Cost: $0.04 flat-rate plan: OpenCode’s price for the tokens, not a charge

- Work: 4 steps 1 session · 3 tool calls, 2 failed

Claude’s note. It wrote nothing. Its web search failed and unpkg answered with a server error; it then got the same package.json from jsDelivr, but replied that a tool had returned an error and it could not complete the request. The run ended there, after four steps and 104 seconds.

Nothing to playnot on the atlas

Jeep Race - WebGPU

A red jeep struck through: MiMo-V2.5-Pro failed the brief — Stuck on “Initializing renderer…”

Task failed. Nothing to drive. The page never gets past “Initializing renderer…”. Its WebGPU renderer is loaded from a path three.js r170 does not have, so it falls back to WebGL2; then the script stops on “Cannot assign to read only property ‘position’ of object”, while it builds the jeep, and the loading screen stays up.

The run OpenCode 2.0.12 · nano-gpt/xiaomi/mimo-v2.5-pro:thinking

- Time: 22 min

- Tokens: 345k 299k read from cache · 26k written

- Cost: $0.03 flat-rate plan: OpenCode’s price for the tokens, not a charge

- Work: 26 steps 1 session · 9 tool calls

Claude’s note. It worked for 22 minutes over 26 steps and wrote one index.html: layered terrain with biomes, instanced trees, rocks and grass, a lake, a spline track with barriers and a start arch, a detailed jeep and three laps over four checkpoints. Twelve times the upstream sent a tool call OpenCode could not parse, and each time it carried on. It tried twice to open the page in OpenCode's browser, but none was connected. Its WebGPU import is a path three.js r170 does not have, so it falls back to WebGL2, and then building the jeep throws on a read-only position: the loading screen never clears. It finished by itself, sure the game was complete.

Nothing to playnot on the atlas

Jeep Race — Three.js WebGPU

A red jeep struck through: Hy3 failed the brief — An error box: “pow is not defined”

Task failed. Nothing to drive. The page opens straight onto an error box that reads “pow is not defined”, and the console shows it thrown while the sky is made, before any world or jeep exists. Nothing else is ever drawn.

The run OpenCode 2.0.12 · nano-gpt/tencent/hy3

- Time: 44 min

- Tokens: 7.32M 6.86M read from cache · 193k written

- Cost: $0.27 flat-rate plan: OpenCode’s price for the tokens, not a charge

- Work: 138 steps 2 sessions · 146 tool calls, 2 failed

Claude’s note. It worked for 44 minutes over 138 steps and wrote fourteen modules: terrain, scenery, a TSL sky and water, a jeep with its own vehicle model, AI rivals, a race, cameras, a HUD and a minimap. It said it could not run a real browser, so it checked the CDN files with curl and every module with node --check, and had a subagent review the code. None of that caught sky.js calling pow without importing it: the page stops at once on 'pow is not defined'. It finished by itself, sure the game was complete.

Nothing to playnot on the atlas

No page written

A red jeep struck through: Kimi K2.6 failed the brief — No files: it stopped at the plan

Task failed. Nothing to drive. It wrote no files. It looked at the empty folder, listed in the chat what the game would have, said it would start by creating the files, and the run ended there, after two steps and 18 seconds.

The run OpenCode 2.0.12 · nano-gpt/moonshotai/kimi-k2.6:thinking

- Time: 18 s

- Tokens: 15k 7k read from cache · 1k written

- Cost: $0.01 flat-rate plan: OpenCode’s price for the tokens, not a charge

- Work: 2 steps 1 session · 1 tool call

Claude’s note. It wrote nothing. It listed the empty folder, set out in the chat what the game would have, said it would start by creating the files, and ended its turn there, after two steps and 18 seconds.

Nothing to playnot on the atlas

Jeep Race - WebGPU Three.js

A red jeep struck through: Ling-3.0-flash failed the brief — START RACE throws at once

Task failed. Nothing to drive. The menu’s START RACE button throws “THREE.WebGPURenderer is not a constructor”: the page loads three.js r160, whose core has no WebGPU renderer, so no scene is ever made and the menu stays up.

The run OpenCode 2.0.12 · nano-gpt/inclusionai/ling-3.0-flash:thinking

- Time: 36 s

- Tokens: 40k 7k read from cache · 14k written

- Cost: $0.00 flat-rate plan: OpenCode’s price for the tokens, not a charge

- Work: 2 steps 1 session · 1 tool call

Claude’s note. It wrote the whole game, one jeep-race/index.html, in a single step and stopped after 36 seconds without opening it. It loads three.js r160 and calls THREE.WebGPURenderer, which that build does not have, so the START RACE button throws and no scene is made.

Nothing to playnot on the atlas

Jeep Safari — Offroad Race

A red jeep struck through: DeepSeek V4 Flash 0731 failed the brief — START RACE does nothing

Task failed. Nothing to drive. The menu shows a START RACE button that does nothing: the game’s script asks unpkg for a WebGPU renderer at a path three.js r170 does not have, the request fails with 404, and the script never runs. The same with WebGPU hidden from the browser.

The run OpenCode 2.0.12 · nano-gpt/deepseek/deepseek-v4-flash-0731:thinking

- Time: 14 min

- Tokens: 416k 323k read from cache · 18k written

- Cost: $0.01 flat-rate plan: OpenCode’s price for the tokens, not a charge

- Work: 17 steps 1 session · 13 tool calls

Claude’s note. It worked for 14 minutes over 17 steps and wrote one jeep-race.html: a menu, terrain on a 300-by-300 grid, a jeep, a track and a HUD. Three provider errors came along the way and it carried on. It served the page and checked it answered with curl; OpenCode's browser was not connected, so it never saw it run. It imports its WebGPU renderer from a path three.js r170 does not ship, so the script never loads and START RACE does nothing. It finished by itself, sure the game was complete.

Nothing to playnot on the atlas

Jeep Race Game - WebGPU

A red jeep struck through: Qwen3.5 397B A17B failed the brief — A HUD over a white screen

Task failed. Nothing to drive. The page shows its HUD, speed 0 km/h and checkpoint 0/0, over a blank white screen, and the console shows “(void 0) is not a constructor” thrown where the renderer is made: the three.js build it loads has no WebGPU renderer at the name it uses. No scene is ever drawn.

The run OpenCode 2.0.12 · nano-gpt/qwen/qwen3.5-397b-a17b:thinking

- Time: 2 min

- Tokens: 109k 0 read from cache · 10k written

- Cost: $0.09 flat-rate plan: OpenCode’s price for the tokens, not a charge

- Work: 8 steps 1 session · 7 tool calls

Claude’s note. It worked for 2 minutes over 8 steps and wrote a Vite project on three.js 0.170: terrain, a jeep with roll cage and spare tyre, checkpoints and a HUD. It started the dev server and declared the game running without looking at it. It calls THREE.WebGPURenderer on the plain three entry, which has none in that release, so the page shows its HUD over a white screen and nothing is drawn.

Nothing to playnot on the atlas

Jeep Race — WebGPU

A red jeep struck through: GLM-5.2 failed the brief — Stuck on “Initializing WebGPU…”

Task failed. Nothing to drive. The page never gets past “Initializing WebGPU…”. Its import map asks unpkg for three.js r169’s three.tsl.js, a file that release does not have; the request fails with 404 and the game’s script never loads. The same with WebGPU hidden from the browser.

The run OpenCode 2.0.12 · nano-gpt/z-ai/glm-5.2:thinking

- Time: 2 min

- Tokens: 34k 10 read from cache · 12k written

- Cost: $0.03 flat-rate plan: OpenCode’s price for the tokens, not a charge

- Work: 2 steps 1 session · 1 tool call

Claude’s note. It wrote the whole game, one jeep-race.html, in a single step and stopped after 108 seconds without opening it: terrain with warped noise and vertex colours, a Catmull-Rom track, a jeep and a race. Its import map points three/tsl at a file three.js r169 does not ship, so the script never loads and the page stays on 'Initializing WebGPU...'.

Nothing to playnot on the atlas

Jeep Race — WebGPU

A red jeep struck through: MiniMax-M3 failed the brief — Stops on a script error at once

Task failed. Nothing to drive. The page stops at once on a red notice, in Russian, that WebGPU is unavailable and should be enabled in Chrome’s flags, with the real error beneath it: “WebGPURenderer is not defined”. WebGPU works in that browser; no scene is ever made.

The run OpenCode 2.0.12 · nano-gpt/minimax/minimax-m3:thinking

- Time: 19 min

- Tokens: 17.25M 17.03M read from cache · 78k written

- Cost: $1.16 flat-rate plan: OpenCode’s price for the tokens, not a charge

- Work: 209 steps 1 session · 208 tool calls, 13 failed

Claude’s note. It worked for 19 minutes over 209 steps, saved three.js r169 next to the game and split it into nine modules: layered terrain, a shader sky, a jeep with its own physics, scattered forests and rocks, checkpoints and a minimap. It tried to fetch Chrome for puppeteer to see the page, but the bench's proxy refused the download host, so it checked the modules with node --check and grep instead. main.js calls WebGPURenderer without the THREE namespace it imported, so the page stops at once on its own 'WebGPU unavailable' notice. It finished by itself, sure the game worked.

Nothing to play[this model on the atlas →](https://evalmap.ai/#minimax-m3)

Jeep Rush — Off-Road Race (WebGPU)

A red jeep struck through: DeepSeek V4 Pro 0813 failed the brief — A HUD over a blank screen

Task failed. Nothing to see. START ENGINE starts a race: the lap timer runs, the speedometer climbs past 120 km/h, checkpoints are called and the jeeps’ dots move round the minimap. But the 3D view stays the page’s dark background from the first frame, not even the sky it sets is drawn, and the console shows no error. The HUD also opens on “position 4/3”.

The run OpenCode 2.0.12 · nano-gpt/deepseek/deepseek-v4-pro-0813:thinking

- Time: 11 min

- Tokens: 753k 513k read from cache · 49k written

- Cost: $0.35 flat-rate plan: OpenCode’s price for the tokens, not a charge

- Work: 19 steps 1 session · 19 tool calls, 2 failed

Claude’s note. It worked for 11 minutes over 19 steps. Its two web searches failed, so it read the three.js r170 WebGPU build itself, then wrote one index.html of about 1,200 lines with layered terrain, a lake, instanced trees and rocks, and AI rivals on a spline track. It tried to open the page in a browser, found none connected, and settled for static checks in Node. The frame loop moves the jeeps, the HUD and the minimap but never calls renderer.render, so nothing is ever drawn. It finished by itself, sure the game was complete.

Nothing to play[this model on the atlas →](https://evalmap.ai/#deepseek-v4-pro)

RIDGE RUSH — WebGPU Jeep Race

A red jeep struck through: GLM-5.3-Prime failed the brief — Only the HUD: no world is drawn

Task failed. Nothing to see. The menu works and a race starts: the lap timer runs, the speedometer moves and the jeep’s dot travels round the minimap. But the 3D world is never drawn. Merging the pine tree’s parts fails (“mergeGeometries() failed with geometry at index 1”), and every frame after that stops inside the renderer on “Cannot read properties of null (reading ‘boundingSphere’)”, so the screen stays dark behind the HUD. The same happens with WebGPU, with its WebGL2 fallback and at Low quality.

The run OpenCode 2.0.12 · openrouter/z-ai/glm-5.3-prime#max

- Time: 33 min

- Tokens: 20.77M 20.31M read from cache · 185k written

- Cost: $13.77 pay-per-token, as OpenCode priced it

- Work: 122 steps 1 session · 123 tool calls, 1 failed

Claude’s note. It worked for 33 minutes over 122 steps and wrote one self-contained index.html, about 2,000 lines, with a carved track, splat-shaded terrain, a lake and scattered trees. It never opened a browser: it tested the terrain and a pure-pursuit autopilot in Node with three.js and fixed four bugs it found there, but never rendered a frame. In the browser the menu and HUD work, yet merging the pine geometry fails and every frame throws in the renderer, so no world is ever drawn. It finished by itself, sure the game was done.

Nothing to playnot on the atlas

Jeep Race

[A frame from the jeep race written by Qwen3.7 PlusPlay it ▶](https://evalmap.ai/jeep/qwen37-plus-cmd/)

12/100

Rubric v4 · on this ladder that is less than a rudimentary demo, where 100 is a game on the level of GTA VI

- runs6/10
- world6/25
- driving3/20
- race3/15
- look4/15
- presentation4/15

Held to 12: in the test drives the jeep fell out of the world, drove off its edge, came apart or never moved, so the parts above, which add up to 26, do not count in full.

Measured at 59.9 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 45.3 with the processor slowed six times, and it drew with WebGPU.

Steep low-poly mountains, lakes and trees with a route minimap and 12 checkpoints, but no rivals; the jeep climbs any slope at about 105 km/h, and past the map's edge it drives on through an empty white void.

The run Command Code 1.50.1 · commandcode/qwen/qwen3.7-plus

- Time: 3 min

- Tokens: 141k 82k read from cache · 15k written

- Cost: $0.07 at alibaba/qwen3.7-plus list price; paid through a flat-rate plan, so not a charge

- Work: 5 steps 1 session · 4 tool calls

Claude’s note. No effort setting for this model on Command Code; run once at the CLI's default. One HTML file with three.js from a CDN, vendored for the page.

[Play it here](https://evalmap.ai/jeep/qwen37-plus-cmd/)not on the atlas

Dustline Rally

[A frame from the jeep race written by GPT-5.6 LunaPlay it ▶](https://evalmap.ai/jeep/gpt56-luna-max-cmd/)

36/100

Rubric v4 · on this ladder that is a rudimentary demo, where 100 is a game on the level of GTA VI

- runs9/10
- world6/25
- driving4/20
- race7/15
- look5/15
- presentation5/15

Measured at 60.1 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 52.1 with the processor slowed six times, and it drew with WebGPU.

Rolling green hills with pines and rocks, a faint dashed road through start gantries, three rivals, laps and a position readout; the jeep reaches about 96 km/h and stays on its wheels, but the road is hard to tell from the grass and the low-poly look stays plain.

The run Command Code 1.50.1 · commandcode/gpt-5.6-luna#max

- Time: 3 min

- Tokens: 112k 60k read from cache · 22k written

- Cost: $0.04 at openai/gpt-5.6-luna list price; paid through a flat-rate plan, so not a charge

- Work: 4 steps 1 session · 8 tool calls

Claude’s note. Max reasoning effort. A Vite project; built with vite build for the page.

[Play it here](https://evalmap.ai/jeep/gpt56-luna-max-cmd/)not on the atlas

Ridgeline Rally — WebGPU

[A frame from the jeep race written by GPT-5.6 TerraPlay it ▶](https://evalmap.ai/jeep/gpt56-terra-max-cmd/)

33/100

Rubric v4 · on this ladder that is a rudimentary demo, where 100 is a game on the level of GTA VI

- runs9/10
- world6/25
- driving4/20
- race3/15
- look6/15
- presentation5/15

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 43.4 with the processor slowed six times, and it drew with WebGPU.

Green rolling hills and pines in a warm haze, a dark dirt road and a styled HUD with a route map; checkpoints and laps but no rivals, and the jeep reaches 155 km/h but wobbles and rolls on the slopes off the road.

The run Command Code 1.50.1 · commandcode/gpt-5.6-terra#max

- Time: 6 min

- Tokens: 184k 113k read from cache · 37k written

- Cost: $0.55 at openai/gpt-5.6-terra list price; paid through a flat-rate plan, so not a charge

- Work: 6 steps 1 session · 8 tool calls

Claude’s note. GPT-5.6 Terra at max effort in Command Code: 6 minutes, 6 steps, a Vite project whose production build it checked, never opening the page.

[Play it here](https://evalmap.ai/jeep/gpt56-terra-max-cmd/)[this model on the atlas →](https://evalmap.ai/#gpt-5-6-terra)

Jeep Desert Rally - Three.js WebGPU Off-Road

[A frame from the jeep race written by Gemini 3.8 FlashPlay it ▶](https://evalmap.ai/jeep/gemini38-flash-high-cmd/)

12/100

Rubric v4 · on this ladder that is less than a rudimentary demo, where 100 is a game on the level of GTA VI

- runs6/10
- world6/25
- driving2/20
- race7/15
- look4/15
- presentation6/15

Held to 12: in the test drives the jeep fell out of the world, drove off its edge, came apart or never moved, so the parts above, which add up to 31, do not count in full.

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 31.9 with the processor slowed six times, and it drew with WebGPU.

Red low-poly desert hills, a road with a start gantry, three rivals, positions and laps; the jeep never passes 27 km/h, props render black, and driven straight on it leaves the hills for an empty beige void.

The run Command Code 1.50.1 · commandcode/google/gemini-3.8-flash#high

- Time: 6 min

- Tokens: 1.19M 826k read from cache · 37k written

- Cost: $0.44 at google/gemini-3.8-flash list price; paid through a flat-rate plan, so not a charge

- Work: 34 steps 1 session · 33 tool calls, 1 failed

Claude’s note. Gemini 3.8 Flash at high effort, the highest Command Code offers for it: 6 minutes, 34 steps, a Vite project it built but never opened.

[Play it here](https://evalmap.ai/jeep/gemini38-flash-high-cmd/)[this model on the atlas →](https://evalmap.ai/#gemini-3-8-flash)

Canyon Run — WebGPU Jeep Race

[A frame from the jeep race written by Claude Opus 5Play it ▶](https://evalmap.ai/jeep/claude-opus-5-max-cmd/)

39/100

Rubric v4 · on this ladder that is a rudimentary demo, where 100 is a game on the level of GTA VI

- runs6/10
- world9/25
- driving5/20
- race7/15
- look7/15
- presentation5/15

Measured at 58.3 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 14 with the processor slowed six times, and it drew with WebGPU.

A gorge cut through layered sandstone with the road on its floor, cacti, rocks and three rival jeeps, gears and 4WD on the HUD, laps, positions, respawn and pause; the dust never draws, as its particle shader fails to compile, and off the road the jeep climbs the canyon walls and tips.

The run Command Code 1.50.1 · commandcode/claude-opus-5#max

- Time: 1 h 50 min

- Tokens: 52.88M 52.10M read from cache · 409k written

- Cost: $38.61 at anthropic/claude-opus-5 list price; paid through a flat-rate plan, so not a charge

- Work: 299 steps 1 session · 300 tool calls

Claude’s note. Claude Opus 5 at max effort in Command Code: 110 minutes, 299 steps, a Vite project it opened and drove again and again in the box's headless Chromium, tuning the canyon's rock strata from its own screenshots.

[Play it here](https://evalmap.ai/jeep/claude-opus-5-max-cmd/)not on the atlas

3D WebGPU Jeep Racing Game

[A frame from the jeep race written by Gemini 3.6 FlashPlay it ▶](https://evalmap.ai/jeep/gemini36-flash-cmd/)

12/100

Rubric v4 · on this ladder that is less than a rudimentary demo, where 100 is a game on the level of GTA VI

- runs6/10
- world6/25
- driving2/20
- race4/15
- look5/15
- presentation5/15

Held to 12: in the test drives the jeep fell out of the world, drove off its edge, came apart or never moved, so the parts above, which add up to 28, do not count in full.

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 36.5 with the processor slowed six times, and it drew with WebGPU.

Grassy hills with pines and rocks, eight checkpoints and a timer; the jeep tumbles over the first rises at 26 km/h and, past the terrain's edge, falls into an empty sky for the rest of the minute.

The run Command Code 1.50.1 · commandcode/google/gemini-3.6-flash

- Time: 2 min

- Tokens: 421k 304k read from cache · 14k written

- Cost: $0.15 at google/gemini-3.6-flash list price; paid through a flat-rate plan, so not a charge

- Work: 18 steps 1 session · 17 tool calls

Claude’s note. Gemini 3.6 Flash at Command Code's default effort, as it refused max: 2 minutes, 18 steps, a Vite project it built but never opened.

[Play it here](https://evalmap.ai/jeep/gemini36-flash-cmd/)not on the atlas

Jeep Race — WebGPU Landscape

[A frame from the jeep race written by InklingPlay it ▶](https://evalmap.ai/jeep/inkling-cmd/)

12/100

Rubric v4 · on this ladder that is less than a rudimentary demo, where 100 is a game on the level of GTA VI

- runs6/10
- world3/25
- driving2/20
- race1/15
- look2/15
- presentation3/15

Held to 12: in the test drives the jeep fell out of the world, drove off its edge, came apart or never moved, so the parts above, which add up to 17, do not count in full.

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 60 with the processor slowed six times, and it drew with WebGL2.

Green hills, a few rocks and pines and one dirt road, drawn with WebGL; no laps, rivals or timer, the jeep holds 45 km/h, and driven straight on it passes the last hill into an empty grey void.

The run Command Code 1.50.1 · commandcode/thinkingmachines/inkling

- Time: 31 s

- Tokens: 73k 52k read from cache · 7k written

- Cost: $0.08 at thinkingmachines/thinkingmachines/Inkling list price; paid through a flat-rate plan, so not a charge

- Work: 4 steps 1 session · 3 tool calls

Claude’s note. Inkling in Command Code, which has no effort setting for it: 31 seconds, 4 steps, one index.html on three.js r160, which has no WebGPU renderer, so the page always takes its WebGL fallback; never opened.

[Play it here](https://evalmap.ai/jeep/inkling-cmd/)[this model on the atlas →](https://evalmap.ai/#inkling)

Jeep Race — WebGPU

[A frame from the jeep race written by Qwen3.7 MaxPlay it ▶](https://evalmap.ai/jeep/qwen37-cmd/)

32/100

Rubric v4 · on this ladder that is a rudimentary demo, where 100 is a game on the level of GTA VI

- runs8/10
- world7/25
- driving2/20
- race6/15
- look4/15
- presentation5/15

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 54.5 with the processor slowed six times, and it drew with WebGPU.

Snowy mountains, lakes and pines around a looped dirt track with rivals and a lap timer; the jeep reaches 198 km/h and rolls over on nearly every slope it meets, and a material error shows in the console.

The run Command Code 1.50.1 · commandcode/qwen/qwen3.7-max

- Time: 2 min

- Tokens: 54k 13k read from cache · 14k written

- Cost: $0.18 at alibaba/qwen3.7-max list price; paid through a flat-rate plan, so not a charge

- Work: 2 steps 1 session · 1 tool call

Claude’s note. Qwen3.7 Max in Command Code, which has no effort setting for it: 2 minutes, 2 steps, one index.html written in a single call with three.js r170 from jsDelivr, never opened.

[Play it here](https://evalmap.ai/jeep/qwen37-cmd/)not on the atlas

Dustline — Canyon Rally

[A frame from the jeep race written by GPT-5.6 SolPlay it ▶](https://evalmap.ai/jeep/gpt56-sol-max-cmd/)

41/100

Rubric v4 · on this ladder that is a solid hobby browser game, where 100 is a game on the level of GTA VI

- runs8/10
- world8/25
- driving4/20
- race7/15
- look7/15
- presentation7/15

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 25.3 with the processor slowed six times, and it drew with WebGPU.

A red-rock canyon of mesas, dunes, boulders and cacti with a coherent look, a named loop with eight checkpoints, rivals, positions and engine sound; off the track the jeep crawls at 41 km/h in second gear.

The run Command Code 1.50.1 · commandcode/gpt-5.6-sol#max

- Time: 8 min

- Tokens: 661k 573k read from cache · 42k written

- Cost: $1.29 at openai/gpt-5.6-sol list price; paid through a flat-rate plan, so not a charge

- Work: 18 steps 1 session · 30 tool calls, 1 failed

Claude’s note. GPT-5.6 Sol at max effort in Command Code: 8 minutes, 18 steps, a Vite project it built and served; it looked for a Chromium on the PATH, did not find the box's own, and checked the page with curl only.

[Play it here](https://evalmap.ai/jeep/gpt56-sol-max-cmd/)not on the atlas

Jeep Canyon Race — Three.js WebGPU

[A frame from the jeep race written by Claude Sonnet 5Play it ▶](https://evalmap.ai/jeep/claude-sonnet-5-max-cmd/)

12/100

Rubric v4 · on this ladder that is less than a rudimentary demo, where 100 is a game on the level of GTA VI

- runs6/10
- world8/25
- driving4/20
- race5/15
- look6/15
- presentation5/15

Held to 12: in the test drives the jeep fell out of the world, drove off its edge, came apart or never moved, so the parts above, which add up to 34, do not count in full.

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 53.6 with the processor slowed six times, and it drew with WebGPU.

Eroded hills with grass, rock and snow, a dirt road with marker posts, three rivals and a lap timer with a finish screen; the jeep holds 166 km/h over any ground, and past the map's edge it drives on through an empty white haze with nothing to stop it.

The run Command Code 1.50.1 · commandcode/claude-sonnet-5#max

- Time: 2 h 2 min

- Tokens: 104.21M 103.32M read from cache · 457k written

- Cost: $26.31 at anthropic/claude-sonnet-5 list price; paid through a flat-rate plan, so not a charge

- Work: 407 steps 1 session · 406 tool calls, 13 failed

Claude’s note. Claude Sonnet 5 at max effort in Command Code: 122 minutes, 407 steps, a Vite project with three.js r186 that it opened and drove in the box's headless Chromium before it stopped.

[Play it here](https://evalmap.ai/jeep/claude-sonnet-5-max-cmd/)not on the atlas

Jeep Rally — three.js WebGPU

[A frame from the jeep race written by Kimi K3Play it ▶](https://evalmap.ai/jeep/kimi-k3-max-cmd/)

Taken with WebGPU hidden: with WebGPU on, this page draws nothing

33/100

Rubric v4 · on this ladder that is a rudimentary demo, where 100 is a game on the level of GTA VI

- runs4/10
- world9/25
- driving4/20
- race5/15
- look6/15
- presentation5/15

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU hidden from the page, so the build took its WebGL2 fallback, paced like a 60 Hz screen; 46 with the processor slowed six times, and it drew with WebGL2.

Scored on its WebGL2 fallback, since on WebGPU it draws nothing: a paved road past lakes, beaches and pine forest under snowy ridges, seven ring checkpoints over timed laps, no rivals; the jeep holds 140 km/h and the GO! banner never clears.

The run Command Code 1.50.1 · commandcode/moonshotai/kimi-k3#max

- Time: 15 min

- Tokens: 514k 398k read from cache · 44k written

- Cost: $0.99 at moonshotai/kimi-k3 list price; paid through a flat-rate plan, so not a charge

- Work: 9 steps 1 session · 11 tool calls

Claude’s note. It worked for 15 minutes over 9 steps, took 0.99 of the subscription's credits, and wrote five files: terrain with lakes, beaches, snowy ridges and pine forest, a paved road with markings and posts, a detailed jeep and seven ring checkpoints over timed laps. It served the files and checked the world code in Node, but never opened the page in a browser. On WebGPU nothing is drawn: an instanced mesh claims two more posts than its buffer holds and Chrome rejects every frame. Its WebGL2 fallback only starts where WebGPU is missing, and there the game runs at 60 fps, though the GO! banner never clears.

[Play it here](https://evalmap.ai/jeep/kimi-k3-max-cmd/)not on the atlas

Canyon Ridge Rally — WebGPU Jeep Racing

[A frame from the jeep race written by Qwen3.8 Max PrimePlay it ▶](https://evalmap.ai/jeep/qwen38-max-prime/)

42/100

Rubric v4 · on this ladder that is a solid hobby browser game, where 100 is a game on the level of GTA VI

- runs8/10
- world10/25
- driving4/20
- race7/15
- look6/15
- presentation7/15

Measured at 59.9 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 29.6 with the processor slowed six times, and it drew with WebGPU.

Snow-capped ridges, pine valleys and a gravel stage carved into the hills, three rivals over three laps and a sector map, at 60 fps on WebGPU; off the first bend the jeep climbs mountainsides at 50 km/h and ends stuck on a snowy one.

[Play it here](https://evalmap.ai/jeep/qwen38-max-prime/)not on the atlas

Jeep Race Game

A red jeep struck through: Ling-3.0-flash-Fin failed the brief — No files: a snippet in the chat

Task failed. Nothing to drive. It wrote no files. In its only step it planned the game, then answered in the chat with a short page to copy: ground of random heights, a flat blue plane for water, a brown box on four cylinders for the jeep and no controls, with three.js r128 from cdnjs and a THREE.WebGPURenderer that build does not have. It closed by asking whether to expand any part of it, and the run ended there.

The run OpenCode 2.0.12 · openrouter/inclusionai/ling-3.0-flash-fin

- Time: 36 s

- Tokens: 8k 384 read from cache · 2k written

- Cost: $0.00 pay-per-token, as OpenCode priced it

- Work: 1 step 1 session · 0 tool calls

Claude’s note. It worked for 36 seconds in a single step and made no tool call. It planned the game in its reasoning and then gave its whole answer there too, a short HTML page in a code block, so OpenCode recorded no reply; nothing was written to disk. It closed by asking whether to expand any part of it, and the run ended there, with OpenCode reporting success. The plumbing check before it, which asks for one file, passed: this model can call tools here.

Nothing to playnot on the atlas

Jeep Race Game - Three.js + WebGPU

A red jeep struck through: Granite 4.2 8B failed the brief — A blue screen that says “Loading”

Task failed. Nothing to drive. Its one script does not load: the browser stops it on “Identifier ‘material’ has already been declared”, so no scene is ever made, and the page shows a plain blue background with a “Loading...” notice, in Russian, in the corner.

The run OpenCode 2.0.12 · openrouter/ibm-granite/granite-4.2-8b#high

- Time: 3 min

- Tokens: 103k 71k read from cache · 13k written

- Cost: $0.01 pay-per-token, as OpenCode priced it

- Work: 6 steps 1 session · 5 tool calls

Claude’s note. It worked for two and a half minutes over six steps. It wrote the whole game into one index.html, loading three.js from jsDelivr, read the file back, named three mistakes in it — an undefined size, a wrong await around rendering, a geometry class three.js no longer ships — made two edits, said it would fix and simplify the rest, and ended its turn there. It never opened the page in a browser, and the page it left does not run.

Nothing to play[this model on the atlas →](https://evalmap.ai/#granite-4-2-8b)

Jeep Race Game

A red jeep struck through: Mercury 2.5 failed the brief — Stuck on “Loading WebGPU…”

Task failed. Nothing to drive. The page never gets past “Loading WebGPU...”: it asks the core three.js build for a WebGPURenderer, which that build does not export, and stops on “THREE.WebGPURenderer is not a constructor”. No terrain and no jeep ever appear.

The run OpenCode 2.0.12 · openrouter/inception/mercury-2.5#high

- Time: 18 s

- Tokens: 46k 0 read from cache · 5k written

- Cost: $0.00 pay-per-token, as OpenCode priced it

- Work: 4 steps 1 session · 3 tool calls, 1 failed

Claude’s note. It finished in 18 seconds over four steps: it wrote index.html and a README in two calls, then tried to preview the page through htmlpreview.github.io at an address it made up, which failed, and stopped there. It never started a server or opened a browser, and its closing summary describes terrain, a road, trees and physics that the page never reaches.

Nothing to playnot on the atlas

Jeep Race — WebGPU — named in Russian by its author; this is the English rendering

A red jeep struck through: Solar Mini 4 failed the brief — Blank screen: nothing is drawn

Task failed. Nothing to drive. The page starts WebGPU and its console reports the terrain, track, trees and rocks built, but the first frame it renders throws “Cannot read properties of undefined (reading ‘constructor’)” inside three.js’s WebGPU renderer, and the game loop stops there. The screen stays empty behind the HUD, and after GO! the speed stays at 0 with the throttle held.

The run OpenCode 2.0.12 · openrouter/upstage/solar-mini4#max

- Time: 1 h 15 min

- Tokens: 41.70M 39.91M read from cache · 225k written

- Cost: $0.32 pay-per-token, as OpenCode priced it

- Work: 223 steps 1 session · 221 tool calls, 4 failed

Claude’s note. It worked on its own for 75 minutes over 223 steps; a connection reset from OpenRouter eight minutes in was retried without a break. It split the game into modules — noise, terrain, track, decor, vehicle — and checked their numbers with a node script: track length, height range, how many trees and rocks. It then installed Playwright and Chromium in its own home and spent most of its last half hour trying to see the page headless; the browser kept closing, and it put that down to the box having no graphics card. It finished with node --check on every file and a README saying the scene should draw in a real browser. It does not; the card above says what happens.

Nothing to playnot on the atlas

Jeep Mountain Race — WebGPU

[A frame from the jeep race written by Ling-3.0-flash-VLPlay it ▶](https://evalmap.ai/jeep/ling3-flash-vl-oc2/)

23/100

Rubric v4 · on this ladder that is a rudimentary demo, where 100 is a game on the level of GTA VI

- runs6/10
- world5/25
- driving2/20
- race3/15
- look3/15
- presentation4/15

Measured at 59.9 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 59.8 with the processor slowed six times, and it drew with WebGL2.

Spiky mountains around a lake with pines and a ring of hoops, after forty seconds of building and on WebGL2, not WebGPU; the jeep is never in view, crawls at 20 km/h and passes no checkpoint in a minute. Laps, a timer and fuel, no rivals.

The run OpenCode 2.0.12 · openrouter/inclusionai/ling-3.0-flash-vl

- Time: 5 min

- Tokens: 1.00M 771k read from cache · 30k written

- Cost: $0.03 pay-per-token, as OpenCode priced it

- Work: 32 steps 1 session · 32 tool calls, 3 failed

Claude’s note. It worked on its own for 5 minutes over 32 steps. It wrote the whole game as one index.html in its first step, taking three.js r164 from unpkg; its two tries at opening the page in a browser failed, so it probed unpkg for the addon files it wanted, reworked the renderer set-up and checked the script with node. On a card the page asks the addons folder for a WebGPURenderer that is not exported there, falls back to WebGL2, and takes about forty seconds to build its world: spiky mountains around a lake, pines, rocks and a ring of checkpoint hoops. The jeep never comes into view and crawls at 20 km/h.

[Play it here](https://evalmap.ai/jeep/ling3-flash-vl-oc2/)not on the atlas

JEEP RACE 4×4 — WebGPU Terrain Rally

[A frame from the jeep race written by DeepSeek-V4-Flash-Vision-ExpPlay it ▶](https://evalmap.ai/jeep/deepseek-v4-flash-vision-exp-max-oc2/)

34/100

Rubric v4 · on this ladder that is a rudimentary demo, where 100 is a game on the level of GTA VI

- runs5/10
- world6/25
- driving5/20
- race7/15
- look4/15
- presentation7/15

Measured at 53.3 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 1.8 with the processor slowed six times, and it drew with WebGPU.

Foggy mountains and pine forest, a start arch, three named rivals on a standings board and ten gates over two laps, after a wait while it generates the world; the jeep holds 100-120 km/h upright, but the pebbled ground reads as rippling water and only posts mark the road.

The run OpenCode 2.0.12 · openrouter/deepseek/deepseek-v4-flash-vision-exp#max

- Time: 1 h 45 min

- Tokens: 54.88M 50.99M read from cache · 297k written

- Cost: $1.34 pay-per-token, as OpenCode priced it

- Work: 242 steps 1 session · 259 tool calls, 2 failed

Claude’s note. It worked on its own for 105 minutes over 242 steps. At 42 minutes OpenRouter dropped the connection once; OpenCode carried on to a finished game, but still exited with code 1. It split the game into modules, with a heightmap in level-of-detail chunks and a road carved into it, installed Chromium in the box, wrote headless tests that go from the menu through a race to the finish, fixed three bugs they found, and copied three.js into vendor/. OpenCode's session export was cut short the first time it was read, so the run was summarised again. On a card: foggy mountains, pine forest, three rivals and ten checkpoint gates, on a pebbled ground that reads as rippling water.

[Play it here](https://evalmap.ai/jeep/deepseek-v4-flash-vision-exp-max-oc2/)not on the atlas

Jeep Race — WebGPU

[A frame from the jeep race written by Qwen3.8-Flash-NextPlay it ▶](https://evalmap.ai/jeep/qwen38-flash-oc2/)

39/100

Rubric v4 · on this ladder that is a rudimentary demo, where 100 is a game on the level of GTA VI

- runs9/10
- world9/25
- driving4/20
- race5/15
- look6/15
- presentation6/15

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 48.8 with the processor slowed six times, and it drew with WebGPU.

Hills under ridged mountains with lakes and pine forest, a dirt road and three laps of ring checkpoints at a clean 60 fps on WebGPU; no rivals, and off the road the jeep crawls at 29 km/h.

The run OpenCode 2.0.12 · openrouter/qwen/qwen3.8-flash

- Time: 56 min

- Tokens: 12.84M 12.45M read from cache · 120k written

- Cost: $0.30 pay-per-token, as OpenCode priced it

- Work: 138 steps 1 session · 137 tool calls, 2 failed

Claude’s note. It worked on its own for 56 minutes over 138 steps. The game was written in the first quarter of an hour; it then checked the world and the physics with a node test whose autopilot drives all three laps, installed Chromium in the box and drove the page on the WebGL2 fallback, as the box has no graphics card, and at the end copied three.js into vendor/ so the page needs no network. On a card it takes the WebGPU path: hills, ridged mountains, lakes and pine forest, a dirt road pressed into the ground, ring checkpoints over three laps, a minimap and an engine sound. There are no rivals.

[Play it here](https://evalmap.ai/jeep/qwen38-flash-oc2/)not on the atlas

Jeep Race Game - Three.js WebGPU

A red jeep struck through: AliceAI-Foundation-80B-A3B-Base failed the brief — Only sky: no ground, no jeep

Task failed. Nothing to drive. Its index.html stops at load on a script error (“Unexpected reserved word” in main.js) and leaves an empty screen. The second page it wrote, game.html, starts WebGPU but draws only sky — no ground, no jeep — while its speed reads “NaN km/h” and the four textures it asks for are Unsplash addresses that do not exist.

The run OpenCode 2.0.12 · selfhosted/aliceai-foundation-80b-a3b-base

- Time: 4 min

- Tokens: 501k 0 read from cache · 27k written

- Cost: none per token the weights ran on a rented GPU, paid by the hour

- Work: 19 steps 1 session · 18 tool calls, 2 failed

Claude’s note. AliceAI-Foundation is a base model with no chat or tool tuning, so the bench reached it through a small adapter: each request was rendered with Yandex’s own agent-trajectory template and sent to vLLM as plain text, and the tool calls the model wrote were handed back to OpenCode as calls. It worked for four and a half minutes over 19 steps: it wrote two near-identical pages, index.html and game.html, loading three.js from a CDN, then an Express server for them, fetched the page once with curl, and stopped there without opening it in a browser. Neither page plays; the card above says what each one shows.

Nothing to playnot on the atlas

DUNE // RUN — WebGPU Jeep Race

[A frame from the jeep race written by Space Bunny AlphaPlay it ▶](https://evalmap.ai/jeep/space-bunny-alpha-max/)

36/100

Rubric v4 · on this ladder that is a rudimentary demo, where 100 is a game on the level of GTA VI

- runs7/10
- world8/25
- driving5/20
- race5/15
- look6/15
- presentation5/15

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 16.7 with the processor slowed six times, and it drew with WebGPU.

A clean 60 fps WebGPU desert circuit with a course the pilot could follow, held back by rivals that never show up and a single biome.

The run OpenCode 2.0.12 · openrouter/stealth/space-bunny-alpha#max

- Time: 10 min

- Tokens: 8.29M 8.14M read from cache · 74k written

- Cost: $0 a free model

- Work: 76 steps 1 session · 88 tool calls, 6 failed

Claude’s note. Second run: the first broke off at step 42 when OpenRouter took the stealth model offline, and the brief was run again from scratch. It worked for 10 minutes on its own, began installing a browser to test the page, stopped that, and checked its code with node and a local server instead. The result is a low-poly desert circuit on WebGPURenderer with mesas, rocks, cacti, a track laid over the dunes, checkpoints, three laps and a polished HUD with a minimap. The two rivals it wrote never appear, and their distance reads NaN; there is no sound.

[Play it here](https://evalmap.ai/jeep/space-bunny-alpha-max/)not on the atlas

DUNE RUN — Rally across the wild

[A frame from the jeep race written by GPT-6 LunaPlay it ▶](https://evalmap.ai/jeep/gpt6luna-max-cx/)

12/100

Rubric v4 · on this ladder that is less than a rudimentary demo, where 100 is a game on the level of GTA VI

- runs6/10
- world4/25
- driving2/20
- race4/15
- look5/15
- presentation7/15

Held to 12: in the test drives the jeep fell out of the world, drove off its edge, came apart or never moved, so the parts above, which add up to 28, do not count in full.

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 39.3 with the processor slowed six times, and it drew with WebGPU.

A styled rally screen over bare sand and one mesa; driven straight on, the jeep runs off the ground into fog at a fixed 104 km/h and never comes back.

The run Codex 0.155.1 · openai/gpt-6-luna#max

- Time: 18 min

- Tokens: 2.02M 1.85M read from cache · 61k written

- Cost: $0.06 at openai/gpt-6-luna list price; paid through a flat-rate plan, so not a charge

- Work: 31 steps 1 session · 26 tool calls, 2 failed

Claude’s note. Finished in 18 minutes in Codex after reading three.js's WebGPU docs on the web; it built the game but never opened it in a browser. A desert rally among buttes and scattered boulders with three timed laps, checkpoints, a minimap, pause and sound, but no rivals, and the chase camera passes through roadside poles.

[Play it here](https://evalmap.ai/jeep/gpt6luna-max-cx/)[this model on the atlas →](https://evalmap.ai/#gpt-6-luna)

Dust Runner — Offroad Racing

[A frame from the jeep race written by GPT-6 SolPlay it ▶](https://evalmap.ai/jeep/gpt6sol-high-cx/)

39/100

Rubric v4 · on this ladder that is a rudimentary demo, where 100 is a game on the level of GTA VI

- runs7/10
- world7/25
- driving5/20
- race7/15
- look6/15
- presentation7/15

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 18 with the processor slowed six times, and it drew with WebGPU.

Red mesas and a sand road with gates, three rivals and a progress bar; it keeps to the road and flags a missed checkpoint, one of the few that plays like a race.

The run Codex 0.155.1 · openai/gpt-6-sol#high

- Time: 13 min

- Tokens: 1.79M 1.68M read from cache · 37k written

- Cost: $0.85 at openai/gpt-6-sol list price; paid through a flat-rate plan, so not a charge

- Work: 43 steps 1 session · 35 tool calls, 2 failed

Claude’s note. Finished in 13 minutes in Codex and checked its own build in headless Chrome, desktop and phone. A point-to-point desert sprint past red mesas, rocks and cacti against three rival jeeps, six checkpoints and nitro, with a polished HUD and sound. The world is thinner than the best here and the course is a single stretch, but it runs smoothly on the fallback.

[Play it here](https://evalmap.ai/jeep/gpt6sol-high-cx/)not on the atlas

Jeep Rally · WebGPU

[A frame from the jeep race written by Claude Opus 5.5Play it ▶](https://evalmap.ai/jeep/opus55-xhigh-cc/)

40/100

Rubric v4 · on this ladder that is a solid hobby browser game, where 100 is a game on the level of GTA VI

- runs5/10
- world10/25
- driving4/20
- race8/15
- look6/15
- presentation7/15

Measured at 46.4 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 20 with the processor slowed six times, and it drew with WebGPU.

A forest with grass, water and five named rivals on a standings board, after ten seconds of generating the world; it runs at 46 fps, and the jeep ends among the trees.

The run Claude Code 2.1.280 · anthropic/claude-opus-5-5#xhigh

- Time: 1 h 37 min

- Tokens: 34.31M 33.54M read from cache · 327k written

- Cost: $16.80 flat-rate plan: Claude Code’s price for the tokens, not a charge

- Work: 132 steps 1 session · 131 tool calls, 4 failed

Claude’s note. Ran for 97 minutes unattended in Claude Code and finished on its own, testing as it went with a headless race simulator it wrote. The result is the richest world in the gallery: an eroded heightmap built in a worker, a river and lake with two fords, forest, grass, a crowd at the start, five rivals, three laps and a menu for quality, laps and a random world. On this GPU-less machine the WebGL2 fallback draws well under a frame a second, so the driving was judged from the code.

[Play it here](https://evalmap.ai/jeep/opus55-xhigh-cc/)[this model on the atlas →](https://evalmap.ai/#claude-opus-5-5-xhigh)

Jeep Rally — three.js WebGPU

[A frame from the jeep race written by GLM-5.3Play it ▶](https://evalmap.ai/jeep/glm53-max-oc2/)

33/100

Rubric v4 · on this ladder that is a rudimentary demo, where 100 is a game on the level of GTA VI

- runs8/10
- world8/25
- driving2/20
- race4/15
- look5/15
- presentation6/15

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 37.7 with the processor slowed six times, and it drew with WebGPU.

A lake among green mountains and an off-track warning over eleven checkpoints; by the end the jeep lies on its roof at 6 km/h.

The run OpenCode 2.0.12 · openrouter/z-ai/glm-5.3#max

- Time: 13 min

- Tokens: 3.43M 3.13M read from cache · 105k written

- Cost: $0.94 pay-per-token, as OpenCode priced it

- Work: 38 steps 1 session · 52 tool calls

Claude’s note. Finished on its own in 13 minutes, no tool call failed: it fetched three.js r170, wrote nine modules, then, finding OpenCode's browser not connected, set up headless Chrome itself, checked frames pixel by pixel, drove the WebGL2 fallback and scripted a full three-lap race. The build runs as shipped: rolling hills with snow on the rim, lakes, a carved dirt track with ruts, pines, rocks and grass, eleven checkpoints, a minimap and engine sound. Here it reached 89 km/h on the WebGL2 fallback; the WebGPU path cannot be seen on this machine, which has no GPU. Flaws: the dust puffs are sliced flat where they meet the ground, and the minimap is very small.

[Play it here](https://evalmap.ai/jeep/glm53-max-oc2/)[this model on the atlas →](https://evalmap.ai/#glm-5-3)

JEEP RALLY 3D — Dust Valley

[A frame from the jeep race written by MiMo-V2.6-ProPlay it ▶](https://evalmap.ai/jeep/mimo-v26-pro/)

36/100

Rubric v4 · on this ladder that is a rudimentary demo, where 100 is a game on the level of GTA VI

- runs7/10
- world8/25
- driving4/20
- race7/15
- look5/15
- presentation5/15

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 24 with the processor slowed six times, and it drew with WebGPU.

A tarmac road with a centre line, a lake, signs and three rivals; the jeep runs off onto the grass and the game calls Wrong Way.

[Play it here](https://evalmap.ai/jeep/mimo-v26-pro/)[this model on the atlas →](https://evalmap.ai/#mimo-v2-6-pro)

Granite Pass

[A frame from the jeep race written by Fugu MaxPlay it ▶](https://evalmap.ai/jeep/fugu-max/)

12/100

Rubric v4 · on this ladder that is less than a rudimentary demo, where 100 is a game on the level of GTA VI

- runs5/10
- world6/25
- driving1/20
- race5/15
- look3/15
- presentation5/15

Held to 12: in the test drives the jeep fell out of the world, drove off its edge, came apart or never moved, so the parts above, which add up to 25, do not count in full.

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 17 with the processor slowed six times, and it drew with WebGPU.

A snowy pass with rivals, but ten seconds in the physics blows up: the speedometer reads 2.5e+64 km/h and the screen stays blank grey for the rest of the minute.

[Play it here](https://evalmap.ai/jeep/fugu-max/)not on the atlas

WebGPU Jeep Racer

[A frame from the jeep race written by GLM-5.3-FlashXPlay it ▶](https://evalmap.ai/jeep/glm53-flashx/)

12/100

Rubric v4 · on this ladder that is less than a rudimentary demo, where 100 is a game on the level of GTA VI

- runs6/10
- world6/25
- driving4/20
- race3/15
- look4/15
- presentation2/15

Held to 12: in the test drives the jeep fell out of the world, drove off its edge, came apart or never moved, so the parts above, which add up to 25, do not count in full.

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 111.9 with the processor slowed six times, and it drew with WebGPU.

Snow and grass hills dotted with tiny trees, a lap and four checkpoints, and a HUD too small to read; driven straight on, the jeep crosses the snow and drives on into an empty pale sky.

[Play it here](https://evalmap.ai/jeep/glm53-flashx/)not on the atlas

Ridge Rally

[A frame from the jeep race written by Grok 4.7Play it ▶](https://evalmap.ai/jeep/grok47-xhigh/)

35/100

Rubric v4 · on this ladder that is a rudimentary demo, where 100 is a game on the level of GTA VI

- runs9/10
- world7/25
- driving4/20
- race6/15
- look5/15
- presentation4/15

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 51.7 with the processor slowed six times, and it drew with WebGPU.

An alpine dirt road through pine forest under snowy peaks, one rival, five gates and a gap timer; steered along it the jeep holds 100 km/h, but hops onto the slopes and ends against a tree.

[Play it here](https://evalmap.ai/jeep/grok47-xhigh/)[this model on the atlas →](https://evalmap.ai/#grok-4-7)

16–19 Sep

JEEP RACE — WebGPU

[A frame from the jeep race written by Step 5 PreviewPlay it ▶](https://evalmap.ai/jeep/step-5-preview-stepfun/)

31/100

Rubric v4 · on this ladder that is a rudimentary demo, where 100 is a game on the level of GTA VI

- runs7/10
- world6/25
- driving3/20
- race7/15
- look3/15
- presentation5/15

Measured at 59 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 21 with the processor slowed six times, and it drew with WebGPU.

Five rivals, a speedometer and a respawn; jagged dark peaks and muddy shading, and on the course the jeep rolls over the steep hills and rights itself again and again.

[Play it here](https://evalmap.ai/jeep/step-5-preview-stepfun/)[this model on the atlas →](https://evalmap.ai/#step-5)

DUST RUN — Desert Rally

[A frame from the jeep race written by Pareto 26.9Play it ▶](https://evalmap.ai/jeep/union-alpha-092026/)

34/100

Rubric v4 · on this ladder that is a rudimentary demo, where 100 is a game on the level of GTA VI

- runs6/10
- world7/25
- driving5/20
- race5/15
- look6/15
- presentation5/15

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 11.5 with the processor slowed six times, and it drew with WebGPU.

A low-poly desert of mesas and cacti in a styled rally screen; steered along its posts, the jeep holds 80 km/h through six of eight checkpoints in the minute, with no rival in sight.

[Play it here](https://evalmap.ai/jeep/union-alpha-092026/)not on the atlas

8–11 Sep

Jeep Valley Rally — WebGPU

[A frame from the jeep race written by Kimi K2.8 PreviewPlay it ▶](https://evalmap.ai/jeep/kimi-k28preview/)

12/100

Rubric v4 · on this ladder that is less than a rudimentary demo, where 100 is a game on the level of GTA VI

- runs6/10
- world6/25
- driving4/20
- race2/15
- look4/15
- presentation2/15

Held to 12: in the test drives the jeep fell out of the world, drove off its edge, came apart or never moved, so the parts above, which add up to 24, do not count in full.

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 46.2 with the processor slowed six times, and it drew with WebGPU.

Low-poly meadows and trees with a pale road the jeep follows at 50-80 km/h, a lap clock and little else; driven straight on, it leaves the meadows within 20 seconds for an empty white void.

[Play it here](https://evalmap.ai/jeep/kimi-k28preview/)not on the atlas

Jeep Race — three.js WebGPU

[A frame from the jeep race written by Devin SWE-2Play it ▶](https://evalmap.ai/jeep/devin-swe2/)

31/100

Rubric v4 · on this ladder that is a rudimentary demo, where 100 is a game on the level of GTA VI

- runs9/10
- world8/25
- driving3/20
- race3/15
- look5/15
- presentation3/15

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 40.5 with the processor slowed six times, and it drew with WebGPU.

Textured relief with peaks and pines, but the jeep reads Airborne the whole drive and the camera ends up inside it; a lap clock and checkpoints.

[Play it here](https://evalmap.ai/jeep/devin-swe2/)not on the atlas

Dust Rally — Jeep Race

[A frame from the jeep race written by MiMo-X-Pro-PreviewPlay it ▶](https://evalmap.ai/jeep/mimo-x-pro/)

37/100

Rubric v4 · on this ladder that is a rudimentary demo, where 100 is a game on the level of GTA VI

- runs9/10
- world7/25
- driving5/20
- race6/15
- look5/15
- presentation5/15

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 46.8 with the processor slowed six times, and it drew with WebGPU.

A snowy valley with pine forest and a dirt course; steered, the jeep laps it in 40 s at 60-80 km/h through checkpoint gates, with rivals, a position and lap and best-lap times.

[Play it here](https://evalmap.ai/jeep/mimo-x-pro/)not on the atlas

JEEP RALLY — WebGPU

[A frame from the jeep race written by DeepSeek V4.1 FlashPlay it ▶](https://evalmap.ai/jeep/deepseek-v4-1-flash-beta/)

12/100

Rubric v4 · on this ladder that is less than a rudimentary demo, where 100 is a game on the level of GTA VI

- runs6/10
- world8/25
- driving4/20
- race4/15
- look5/15
- presentation5/15

Held to 12: in the test drives the jeep fell out of the world, drove off its edge, came apart or never moved, so the parts above, which add up to 32, do not count in full.

Measured at 59.7 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 24.5 with the processor slowed six times, and it drew with WebGPU.

Green hills and mountains with a checkpoint strip; off the first bend the jeep drove into a lake, through it and under the world, and spent the rest of the minute beneath the terrain.

[Play it here](https://evalmap.ai/jeep/deepseek-v4-1-flash-beta/)not on the atlas

1–5 Sep

Hills & Jeeps — jeep race / WebGPU — named in Russian by its author; this is the English rendering

[A frame from the jeep race written by Qwen3.8 27BPlay it ▶](https://evalmap.ai/jeep/cerebrast-qwen38-27b-1300toksec-medium-effort/)

12/100

Rubric v4 · on this ladder that is less than a rudimentary demo, where 100 is a game on the level of GTA VI

- runs6/10
- world4/25
- driving4/20
- race4/15
- look3/15
- presentation4/15

Held to 12: in the test drives the jeep fell out of the world, drove off its edge, came apart or never moved, so the parts above, which add up to 25, do not count in full.

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 55.4 with the processor slowed six times, and it drew with WebGPU.

Bare green and beige hills with specks for props, seven checkpoints and a lap clock; driven straight on, the jeep leaves the hills after a minute and runs on over nothing but a brown haze.

[Play it here](https://evalmap.ai/jeep/cerebrast-qwen38-27b-1300toksec-medium-effort/)[this model on the atlas →](https://evalmap.ai/#qwen3-8-27b)

Jeep Rally — three.js WebGPU

[A frame from the jeep race written by Claude Fable 5.1Play it ▶](https://evalmap.ai/jeep/fable51-medium-effort/)

31/100

Rubric v4 · on this ladder that is a rudimentary demo, where 100 is a game on the level of GTA VI

- runs9/10
- world8/25
- driving3/20
- race3/15
- look5/15
- presentation3/15

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 43.6 with the processor slowed six times, and it drew with WebGPU.

Textured hills, bushes and peaks; the jeep is Airborne all drive and ends with the camera in a rock face. A lap clock, no rivals.

[Play it here](https://evalmap.ai/jeep/fable51-medium-effort/)[this model on the atlas →](https://evalmap.ai/#claude-fable-5-1-medium)

Trail Rush — Alpine Rally

[A frame from the jeep race written by GPT-6 AstraPlay it ▶](https://evalmap.ai/jeep/astra/)

39/100

Rubric v4 · on this ladder that is a rudimentary demo, where 100 is a game on the level of GTA VI

- runs7/10
- world9/25
- driving3/20
- race6/15
- look7/15
- presentation7/15

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 50.6 with the processor slowed six times, and it drew with WebGPU.

The best-drawn forest here, with three rivals, checkpoints and a back-on-the-trail reset; steered, the jeep crawls through bends in wheelspin and tops out near 60 km/h, but it can be kept going.

[Play it here](https://evalmap.ai/jeep/astra/)[this model on the atlas →](https://evalmap.ai/#gpt-6-astra)

Jeep Race 3D — WebGPU Off-Road

[A frame from the jeep race written by Muse Spark 1.3Play it ▶](https://evalmap.ai/jeep/muse13-max-effort/)

35/100

Rubric v4 · on this ladder that is a rudimentary demo, where 100 is a game on the level of GTA VI

- runs9/10
- world8/25
- driving5/20
- race3/15
- look6/15
- presentation4/15

Measured at 59.9 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 45.4 with the processor slowed six times, and it drew with WebGPU.

A lake, and a dirt track lined with cones, tyres and flags under peaks; steered, the jeep holds 60-77 km/h along it, but there are no rivals, only a lap clock.

[Play it here](https://evalmap.ai/jeep/muse13-max-effort/)[this model on the atlas →](https://evalmap.ai/#muse-spark-1-3)

Jeep Race — WebGPU Offroad

[A frame from the jeep race written by Omen AlphaPlay it ▶](https://evalmap.ai/jeep/omen-alpha-maybe-glm5/)

12/100

Rubric v4 · on this ladder that is less than a rudimentary demo, where 100 is a game on the level of GTA VI

- runs6/10
- world7/25
- driving1/20
- race2/15
- look4/15
- presentation4/15

Held to 12: in the test drives the jeep fell out of the world, drove off its edge, came apart or never moved, so the parts above, which add up to 24, do not count in full.

Measured at 59.9 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 59.3 with the processor slowed six times, and it drew with WebGPU.

Low-poly green mountains with no race to speak of; off the road the jeep rolled down a slope and came apart, its wheels lying beside an upside-down body at 0 km/h.

[Play it here](https://evalmap.ai/jeep/omen-alpha-maybe-glm5/)not on the atlas

Game104 — Jeep Race

[A frame from the jeep race written by Muse Spark 1.3Play it ▶](https://evalmap.ai/jeep/muse13/)

38/100

Rubric v4 · on this ladder that is a rudimentary demo, where 100 is a game on the level of GTA VI

- runs9/10
- world9/25
- driving4/20
- race6/15
- look6/15
- presentation4/15

Measured at 59.9 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 50.3 with the processor slowed six times, and it drew with WebGPU.

Snowy peaks, a posted dirt track with checkpoints and three rivals with headlights; steered, the jeep follows it but never passes 37 km/h and drops to fourth.

[Play it here](https://evalmap.ai/jeep/muse13/)[this model on the atlas →](https://evalmap.ai/#muse-spark-1-3)

Jeep Extreme 4x4 - WebGPU AAA Race

[A frame from the jeep race written by Gemini 3.8 FlashPlay it ▶](https://evalmap.ai/jeep/aaa-jeep-game-gemini38flash/)

12/100

Rubric v4 · on this ladder that is less than a rudimentary demo, where 100 is a game on the level of GTA VI

- runs6/10
- world8/25
- driving3/20
- race6/15
- look5/15
- presentation6/15

Held to 12: in the test drives the jeep fell out of the world, drove off its edge, came apart or never moved, so the parts above, which add up to 34, do not count in full.

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 4.1 with the processor slowed six times, and it drew with WebGPU.

Pine hills and three rivals; the jeep sat stuck on the start line in reverse, and after the reset dropped out of the world into an empty blue void for the rest of the minute.

[Play it here](https://evalmap.ai/jeep/aaa-jeep-game-gemini38flash/)[this model on the atlas →](https://evalmap.ai/#gemini-3-8-flash)

Jeep Extreme 4x4 - WebGPU Race

[A frame from the jeep race written by Gemini 3.8 FlashPlay it ▶](https://evalmap.ai/jeep/gemini38flash/)

12/100

Rubric v4 · on this ladder that is less than a rudimentary demo, where 100 is a game on the level of GTA VI

- runs6/10
- world7/25
- driving2/20
- race3/15
- look5/15
- presentation6/15

Held to 12: in the test drives the jeep fell out of the world, drove off its edge, came apart or never moved, so the parts above, which add up to 29, do not count in full.

Measured at 59.9 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 14.4 with the processor slowed six times, and it drew with WebGPU.

Dark pine hills, a clean HUD and one lap clock; driven straight on, the jeep leaves the hills within 25 seconds and runs on alone through an empty sky at 137 km/h.

[Play it here](https://evalmap.ai/jeep/gemini38flash/)[this model on the atlas →](https://evalmap.ai/#gemini-3-8-flash)

Jeep Ridge II — WebGPU off-road race

[A frame from the jeep race written by Claude Fable 5.1Play it ▶](https://evalmap.ai/jeep/fable51/)

32/100

Rubric v4 · on this ladder that is a rudimentary demo, where 100 is a game on the level of GTA VI

- runs6/10
- world10/25
- driving4/20
- race3/15
- look6/15
- presentation3/15

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 4.1 with the processor slowed six times, and it drew with WebGPU.

A lake, a gravel road through forest and snowy peaks — the best-detailed ground here — but the jeep climbs off the road and sticks at 0; one lap clock, no rivals.

[Play it here](https://evalmap.ai/jeep/fable51/)[this model on the atlas →](https://evalmap.ai/#claude-fable-5-1)

Jeep Ridge — WebGPU race

[A frame from the jeep race written by Qwen3.8 Max 0902Play it ▶](https://evalmap.ai/jeep/qwen38-max-0902/)

24/100

Rubric v4 · on this ladder that is a rudimentary demo, where 100 is a game on the level of GTA VI

- runs9/10
- world4/25
- driving2/20
- race3/15
- look3/15
- presentation3/15

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 42.2 with the processor slowed six times, and it drew with WebGPU.

Cone pines on a grey hill by a flat lake; the jeep tumbles off the start and runs at a fixed 122 km/h.

[Play it here](https://evalmap.ai/jeep/qwen38-max-0902/)[this model on the atlas →](https://evalmap.ai/#qwen3-8-max)

Jeep Rally — Off-Road Rage | Game104

[A frame from the jeep race written by Muse Spark 1.2Play it ▶](https://evalmap.ai/jeep/muse12/)

30/100

Rubric v4 · on this ladder that is a rudimentary demo, where 100 is a game on the level of GTA VI

- runs8/10
- world3/25
- driving4/20
- race7/15
- look3/15
- presentation5/15

Measured at 59.9 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 57 with the processor slowed six times, and it drew with WebGL2.

Almost bare ground and a flat sea, but three rivals on a standings board and eight checkpoints.

[Play it here](https://evalmap.ai/jeep/muse12/)not on the atlas

Ridgeline Rally — WebGPU off-road

[A frame from the jeep race written by Qwen3.8 FlashPlay it ▶](https://evalmap.ai/jeep/qwen38flash-local-2xsparks/)

29/100

Rubric v4 · on this ladder that is a rudimentary demo, where 100 is a game on the level of GTA VI

- runs4/10
- world7/25
- driving4/20
- race3/15
- look4/15
- presentation7/15

Measured at 59.9 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 1.3 with the processor slowed six times, and it drew with WebGPU.

A thorough menu, then thirty seconds before its start works; dark noisy hills, and a lap counter that counts a lap every two seconds.

[Play it here](https://evalmap.ai/jeep/qwen38flash-local-2xsparks/)not on the atlas

26–28 Aug

HY4 DUST — Jeep Rally (WebGPU)

[A frame from the jeep race written by HY4 PreviewPlay it ▶](https://evalmap.ai/jeep/tecent-hy4-preview/)

35/100

Rubric v4 · on this ladder that is a rudimentary demo, where 100 is a game on the level of GTA VI

- runs6/10
- world8/25
- driving3/20
- race7/15
- look6/15
- presentation5/15

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 10.6 with the processor slowed six times, and it drew with WebGPU.

A mountain road past forest and banners with three named rivals, ending airborne on a snowfield in last place.

[Play it here](https://evalmap.ai/jeep/tecent-hy4-preview/)not on the atlas

Pine Ridge Rally — WebGPU offroad racer

[A frame from the jeep race written by GLM-5.3-FlashPlay it ▶](https://evalmap.ai/jeep/glm53-flash-2x-sparks-localai/)

33/100

Rubric v4 · on this ladder that is a rudimentary demo, where 100 is a game on the level of GTA VI

- runs7/10
- world8/25
- driving2/20
- race6/15
- look5/15
- presentation5/15

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 21.7 with the processor slowed six times, and it drew with WebGPU.

A lake, dunes and pines; the field piles up at the start and the jeep ends on its side on a dune. Laps, positions, a good minimap.

[Play it here](https://evalmap.ai/jeep/glm53-flash-2x-sparks-localai/)[this model on the atlas →](https://evalmap.ai/#glm-5-3-flash)

Jeep Race · WebGPU

[A frame from the jeep race written by Qwen3.8 27BPlay it ▶](https://evalmap.ai/jeep/qwen38-27b-local-4090rtx/)

27/100

Rubric v4 · on this ladder that is a rudimentary demo, where 100 is a game on the level of GTA VI

- runs8/10
- world5/25
- driving3/20
- race3/15
- look4/15
- presentation4/15

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 38.4 with the processor slowed six times, and it drew with WebGPU.

Pines on a meadow, then a bare grey plain where the jeep stops at 2 km/h; a lap clock and a record.

[Play it here](https://evalmap.ai/jeep/qwen38-27b-local-4090rtx/)[this model on the atlas →](https://evalmap.ai/#qwen3-8-27b)

DUST PEAKS RALLY — WebGPU Jeep Racing

[A frame from the jeep race written by GLM-5.3-FlashPlay it ▶](https://evalmap.ai/jeep/glm5-3-flash/)

31/100

Rubric v4 · on this ladder that is a rudimentary demo, where 100 is a game on the level of GTA VI

- runs7/10
- world6/25
- driving3/20
- race6/15
- look4/15
- presentation5/15

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 24.8 with the processor slowed six times, and it drew with WebGPU.

A meadow road with three named rivals on the board; off the road the jeep tumbles across an empty green hill.

[Play it here](https://evalmap.ai/jeep/glm5-3-flash/)[this model on the atlas →](https://evalmap.ai/#glm-5-3-flash)

Jeep Rally — WebGPU

[A frame from the jeep race written by GLM-5.3Play it ▶](https://evalmap.ai/jeep/glm53/)

32/100

Rubric v4 · on this ladder that is a rudimentary demo, where 100 is a game on the level of GTA VI

- runs8/10
- world7/25
- driving2/20
- race5/15
- look5/15
- presentation5/15

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 27.5 with the processor slowed six times, and it drew with WebGPU.

A dirt road between rocky peaks with three rivals ahead; the jeep ends on its side against a boulder at 0.

[Play it here](https://evalmap.ai/jeep/glm53/)[this model on the atlas →](https://evalmap.ai/#glm-5-3)

Jeep Off-Road Rally 4x4 | WebGPU

[A frame from the jeep race written by Gemini 3.7 FlashPlay it ▶](https://evalmap.ai/jeep/gemini-3-7-flash/)

12/100

Rubric v4 · on this ladder that is less than a rudimentary demo, where 100 is a game on the level of GTA VI

- runs6/10
- world4/25
- driving2/20
- race6/15
- look4/15
- presentation6/15

Held to 12: in the test drives the jeep fell out of the world, drove off its edge, came apart or never moved, so the parts above, which add up to 28, do not count in full.

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 14.7 with the processor slowed six times, and it drew with WebGPU.

Bare dunes with rivals, positions and a tidy HUD; driven straight on, the jeep leaves the world within 20 seconds and hangs in empty white at 137 km/h.

[Play it here](https://evalmap.ai/jeep/gemini-3-7-flash/)not on the atlas

18 Aug

Jeep Race — WebGPU

[A frame from the jeep race written by Model not recordedPlay it ▶](https://evalmap.ai/jeep/jeep-race-webgpu/)

24/100

Rubric v4 · on this ladder that is a rudimentary demo, where 100 is a game on the level of GTA VI

- runs7/10
- world4/25
- driving4/20
- race3/15
- look3/15
- presentation3/15

Measured at 59.9 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 23.1 with the processor slowed six times, and it drew with WebGPU.

A grass plain with a few pines and cones; the jeep holds 60 km/h into an empty fogged hill, alone.

[Play it here](https://evalmap.ai/jeep/jeep-race-webgpu/)not on the atlas

14 Aug

MudTrail 4x4: WebGPU Off-Road Championship

[A frame from the jeep race written by Model not recordedPlay it ▶](https://evalmap.ai/jeep/jeep-webgpu-race/)

12/100

Rubric v4 · on this ladder that is less than a rudimentary demo, where 100 is a game on the level of GTA VI

- runs6/10
- world6/25
- driving2/20
- race5/15
- look3/15
- presentation5/15

Held to 12: in the test drives the jeep fell out of the world, drove off its edge, came apart or never moved, so the parts above, which add up to 27, do not count in full.

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 13.3 with the processor slowed six times, and it drew with WebGPU.

Canyon rock that breaks into loose shards and two tracks with checkpoints; in a minute's drive the jeep was thrown into an empty peach sky and hung there until the reset, then stuck in a slope.

[Play it here](https://evalmap.ai/jeep/jeep-webgpu-race/)not on the atlas

3 Aug

Jeep Race — WebGPU

[A frame from the jeep race written by Model not recordedPlay it ▶](https://evalmap.ai/jeep/jeep-race/)

32/100

Rubric v4 · on this ladder that is a rudimentary demo, where 100 is a game on the level of GTA VI

- runs7/10
- world7/25
- driving4/20
- race5/15
- look5/15
- presentation4/15

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 22.7 with the processor slowed six times, and it drew with WebGPU.

Pines on rolling hills, gates with light beams and three rivals; the jeep runs wide onto the grass and crawls.

[Play it here](https://evalmap.ai/jeep/jeep-race/)not on the atlas

23–24 Jul

DUST RAIDERS — jeep racing — named in Russian by its author; this is the English rendering

[A frame from the jeep race written by Model not recordedPlay it ▶](https://evalmap.ai/jeep/dust-raiders/)

20/100

Rubric v4 · on this ladder that is a rudimentary demo, where 100 is a game on the level of GTA VI

- runs6/10
- world7/25
- driving4/20
- race7/15
- look5/15
- presentation6/15

Held to 20: in the test drives the jeep flipped, or wedged itself on the course so that backing off did not free it, so the parts above, which add up to 35, do not count in full.

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 28.6 with the processor slowed six times, and it drew with WebGL2.

Dunes, cacti and five named rivals on a live board; steered along its road the jeep can run a clean minute at 60-130 km/h, but in two drives of three it rolled or wedged at the course's edge and stayed.

[Play it here](https://evalmap.ai/jeep/dust-raiders/)not on the atlas

DESERT RAID — named in Russian by its author; this is the English rendering

[A frame from the jeep race written by Model not recordedPlay it ▶](https://evalmap.ai/jeep/desert-jeep-rally/)

23/100

Rubric v4 · on this ladder that is a rudimentary demo, where 100 is a game on the level of GTA VI

- runs8/10
- world3/25
- driving3/20
- race4/15
- look2/15
- presentation3/15

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 56.1 with the processor slowed six times, and it drew with WebGL2.

Flat sand, a few white rocks and a fog glare that swallows the frame; the jeep ends stuck at 0 in third of three.

[Play it here](https://evalmap.ai/jeep/desert-jeep-rally/)not on the atlas

16–19 Jul

Jeep Road Runner — C64

[A frame from the jeep race written by Model not recordedPlay it ▶](https://evalmap.ai/jeep/jeep-road-runner/)

29/100

Rubric v4 · on this ladder that is a rudimentary demo, where 100 is a game on the level of GTA VI

- runs8/10
- world3/25
- driving5/20
- race3/15
- look4/15
- presentation6/15

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 56.6 with the processor slowed six times, and it drew with a 2D canvas.

A lane runner, not a race: it overtakes, burns fuel and ends Wrecked after 1.5 km with a score and a record, in a retro CRT look.

[Play it here](https://evalmap.ai/jeep/jeep-road-runner/)not on the atlas

Jeep Rally — WebGPU

[A frame from the jeep race written by Model not recordedPlay it ▶](https://evalmap.ai/jeep/jeep-rally/)

27/100

Rubric v4 · on this ladder that is a rudimentary demo, where 100 is a game on the level of GTA VI

- runs7/10
- world5/25
- driving4/20
- race4/15
- look3/15
- presentation4/15

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 22.3 with the processor slowed six times, and it drew with WebGPU.

Grass hills with a mud track and ring checkpoints; steered, the jeep holds 69 km/h along it, but rain streaks and a flat blue sheet cut across the whole frame.

[Play it here](https://evalmap.ai/jeep/jeep-rally/)not on the atlas

Desert Rally — jeep racing — named in Russian by its author; this is the English rendering

[A frame from the jeep race written by Model not recordedPlay it ▶](https://evalmap.ai/jeep/jeep-racing/)

25/100

Rubric v4 · on this ladder that is a rudimentary demo, where 100 is a game on the level of GTA VI

- runs9/10
- world2/25
- driving4/20
- race5/15
- look2/15
- presentation3/15

Measured at 59.9 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 54.7 with the processor slowed six times, and it drew with WebGPU.

A flat sand plane marked out with posts and a box for a jeep; steered, it laps in about 45 s at 100 km/h against three rivals and runs fourth of four.

[Play it here](https://evalmap.ai/jeep/jeep-racing/)not on the atlas

Jeep Racing · Three.js WebGPU — named in Russian by its author; this is the English rendering

[A frame from the jeep race written by Model not recordedPlay it ▶](https://evalmap.ai/jeep/jeep-rally-webgpu/)

28/100

Rubric v4 · on this ladder that is a rudimentary demo, where 100 is a game on the level of GTA VI

- runs9/10
- world5/25
- driving2/20
- race5/15
- look3/15
- presentation4/15

Measured at 60 frames a second in a 45-second drive, 1280×720, on an RTX 5090 in Chrome with WebGPU on, paced like a 60 Hz screen; 48.7 with the processor slowed six times, and it drew with WebGPU.

Grey-green hills and a rival at the start, then the jeep sits stuck at 1 km/h on a bare slope.

[Play it here](https://evalmap.ai/jeep/jeep-rally-webgpu/)not on the atlas

## How to read this

Where the names come from. Each of the older builds lives in its own project named after the model that wrote it, and that name is the only record of who wrote what. Since 21 September a build comes as a project folder instead, with the word of the person who ran it for the model and the effort that wrote it. A build from EvalMap’s own bench names the model its session log records. Where the maker’s own name for the model differs, the card uses the maker’s: the stealth model Union Alpha was revealed as Unbiased’s Pareto 26.9, a blend of several models rather than one, and the build whose project says MiMo X Pro was written by what Xiaomi calls MiMo-X-Pro-Preview, a preview offered only inside its MiMo Desktop app. Omen Alpha is still a codename that no maker has claimed; its card repeats only the guess in its project’s name. The 9 builds older than the series say nothing in their project names about who wrote them, so their cards say the model was not recorded rather than guessing. Where a build states a reasoning effort or an unusual way of serving the model — a local GPU, a fast host — the card states it too.

Running here. The builds are shown here and nowhere else, served from this domain. The older ones are copies of pages their builders published themselves, and three kinds of thing were changed in them. Addresses: files that used to be fetched from a CDN or from Google Fonts are saved beside the build and pointed at locally, so nothing a build asks for is loaded from anywhere else. On-screen Russian, put into English as “In English” below describes. And a strip along the top, added to every build alike: it names the model that wrote the build, the day it went up and the way back here, because a build opened on its own says none of that. It is one block of plain markup and one stylesheet from this domain, with no script behind it and nothing kept in the browser. It moves the build down by its own height rather than covering it, so a display pinned to the top of the window sits just below the strip and controls pinned to the bottom stay on the bottom edge; on a touch screen one build, whose controls would otherwise lose their lower edge, keeps its place and the strip lies over its top. The word “hide” folds the strip away and gives the build the whole window until the page is loaded again. Since 21 September a build is published nowhere but here: it is built from its own source with its own Vite config, relative file paths the one setting added, and nothing in it is changed but that strip.

What the dates mean. The date on a card is the day that build was published, not how long the model took and not the day its maker released it. The earlier batch captures are dated 23 September – 3 October 2026; individually documented captures give their date and method on the card.

What the pictures are. Standard captures are made by opening the copy published here in a headless browser, start the run, hold the accelerator for a few seconds, save the picture. The frames come from these copies rather than from the originals, so a tile shows what this page actually serves, English text and all. These builds ask for WebGPU, so standard captures use an RTX 5090 in Chrome with WebGPU on and frames paced like a 60 Hz screen. Individually annotated pilot or software-browser frames may use another renderer and are identified on their cards; they are not native-GPU performance measurements. A prettier still is not a better model.

In English. 31 of the builds wrote their menus, timers and loading lines in Russian, and 10 named themselves in it. The copies here were put into English from a table of 832 phrases, by hand: the tool replaces text a browser shows — text in the HTML, the title, alt, placeholder and aria-label attributes, and the string literals the scripts draw their words from — and touches nothing else, so the comments, the names and the logic are still exactly what the model wrote. Every replacement is listed in sources/jeep-english.json, line by line, and the build each one landed in is counted in sources/jeep-english-log.json.

The bench. Since 22 September a build can also come from EvalMap’s own bench. The brief goes to the model it names exactly as the owner writes it — English, with one Russian phrase asking for a detailed landscape — and is never reworded. It runs unattended in OpenCode, or in Claude Code, Codex or Command Code when the owner names one, in a new folder and a clean harness: none of the owner’s plugins, settings, extra tools or skills, and the tool that asks a person a question switched off, because an unattended run cannot answer it. Whatever the model hands back is published as it is; nobody fixes or finishes it. The card then says what the run took — time on the clock, tokens and dollars, from the harness’s own session records with any subagents it started counted in — and carries a short note by Claude, the agent that runs the bench, on how the run went and what came of it. The dollars are the harness’s pricing of the tokens, or the maker’s list price where the harness prices nothing: on a flat-rate plan or a subscription that is what they would cost, not what was charged. 77 of the builds came this way.

The score. Completed scoring reviews give cards a number from 0 to 100, shown on the board at the top of this page. It is scored against one written rubric where 100 is a finished open-world driving game on the level of GTA VI — an absolute scale, not a curve over these builds, so a game written from one line of brief belongs in the thirties and anything over 60 would be a game a studio could sell. Six parts carry it: runs (10), world (25), because the brief asks for the landscape to be detailed, driving (20), race (15), look (15) and presentation (15). Each part is rated on the same ladder — rudimentary, a solid hobby browser game, a good commercial indie game, an AA production, AAA — and every scored card shows all six, so the total can be argued with part by part. Scored builds are looked at the same way and none is changed: it is started in a browser here, driven for half a minute or longer, and read where a picture cannot show whether a thing is there. It is one reader’s judgement, written down: Claude’s, the same agent that runs the bench. The rubric is tools/jeepbench/RUBRIC.md; when it changes its version changes, and scores from two versions are not compared. 83 of the 142 builds carry a score. 4 do not have a completed scoring review; each card explains why where its score would be.

What it is not. No build was retried for a better result — the one bench run made twice, Space Bunny Alpha’s, was cut off when its provider went offline, and its card says so — and nothing in how one plays was fixed, or timed. The score is a judgement, not a measurement: one reader against a written ladder, with the six parts shown so the number can be argued with. It is given to the copy published here, which is all these builds have in common — so it ranks builds, not models under equal conditions. Only a bench build has a record of the harness it was written in, the effort it ran at and the brief it was given word for word; for an older build none of that is known, and two builds by the same model are only loosely comparable. The bench’s time, tokens and dollars say what a run cost, not how good its build is, and no number on the atlas comes from them.

The atlas links. 51 of the 142 cards name a model that is also among the configurations on the atlas, and link to the row there. The rest do not: a codename, preview releases, an agent, a blend of several models, a model none of the atlas’s sources covers, and the older builds whose author is unknown. A card links to its model, which the published benchmarks may have measured at a different reasoning effort than the build used.

Nothing that identifies you. This page keeps nothing in your browser but a light or dark theme you pin yourself with the switch in the top bar, and loads no third-party script of its own and no web font. Every file a build asks for comes from this domain: the CDN copies of three.js and the web fonts two builds used are saved here beside them. The exception is the one the atlas discloses too — Cloudflare’s zone adds its cookieless page counter (cloudflareinsights.com) to every HTML page on this site, these builds included; it sets no cookie and reports totals only. The builds are other people’s code rather than this site’s: they draw to a canvas and some of them remember settings or best runs in the browser’s own storage, and none of them sends anything anywhere.

EVALMAP · JEEP BUILDSContact: [support@evalmap.ai](mailto:support@evalmap.ai)No ads, no cookies, no accounts.Builds published 16 Jul 2026 – 7 Nov 2026 · frames captured 23 September – 3 October 2026.[back to the atlas](https://evalmap.ai/) · [sources & licences](https://evalmap.ai/licenses/) · [what changed](https://evalmap.ai/changelog/) · [small tests](https://evalmap.ai/tests/) · [EvalMapBench](https://evalmap.ai/evalmapbench/)