# EvalMap

> Compare 88 AI model configurations in 3D: coding, reasoning and arena benchmarks against speed and price per million tokens. Every value names its source.

Artificial Analysis snapshot of 4 November 2026. The full data: https://evalmap.ai/llms-full.txt and https://evalmap.ai/data.json.

## Highest Terminal-Bench 4.0 scores in the 4 November 2026 Artificial Analysis snapshot

1. [Kimi K3.1 Pro](https://evalmap.ai/#kimi-k3-1-pro): 67.2%
2. [Gemini 4 Carbon (High)](https://evalmap.ai/#gemini-4-carbon): 65.2%
3. [Claude Sonnet 5.5 (max)](https://evalmap.ai/#claude-sonnet-5-5): 63.6%
4. [Gemini 4 Carbon (Low)](https://evalmap.ai/#gemini-4-carbon-low): 62.6%
5. [Gemini 4 Carbon (Medium)](https://evalmap.ai/#gemini-4-carbon-medium): 60.6%
6. [Claude Opus 5.5 (max)](https://evalmap.ai/#claude-opus-5-5): 59.6%
7. [Claude Opus 5.5 (xhigh)](https://evalmap.ai/#claude-opus-5-5-xhigh): 59.6%
8. [GPT-6 Astra (xhigh)](https://evalmap.ai/#gpt-6-astra-xhigh): 59.6%
9. [GPT-6 Astra (max)](https://evalmap.ai/#gpt-6-astra): 59.1%
10. [Claude Sonnet 5.5 (xhigh)](https://evalmap.ai/#claude-sonnet-5-5-xhigh): 57.1%

## Pages

- [Sources and licences](https://evalmap.ai/licenses/) · Markdown: https://evalmap.ai/licenses/index.md
- [Model releases](https://evalmap.ai/releases/) · Markdown: https://evalmap.ai/releases/index.md
- [Jeep gallery](https://evalmap.ai/jeep/) · Markdown: https://evalmap.ai/jeep/index.md
- [Changelog](https://evalmap.ai/changelog/) · Markdown: https://evalmap.ai/changelog/index.md
- [Small tests](https://evalmap.ai/tests/) · Markdown: https://evalmap.ai/tests/index.md
- [EvalMapBench](https://evalmap.ai/evalmapbench/) · Markdown: https://evalmap.ai/evalmapbench/index.md

## Sources

- Artificial Analysis (https://artificialanalysis.ai/): as of 2026-11-04; licence Proprietary · taken through the AA Data API (Pro), with attribution (https://artificialanalysis.ai/terms-of-use)
- Terminal-Bench 4.0 official board (https://www.tbench.ai/leaderboard): as of 2026-09-02; licence Apache-2.0 (https://github.com/harbor-framework/terminal-bench/blob/main/LICENSE)
- DeepSWE v1.1 (https://deepswe.datacurve.ai/): as of 2026-09-22; licence Apache-2.0 (https://github.com/datacurve-ai/deep-swe/blob/main/LICENSE)
- EvalMapBench 6.7 (https://evalmap.ai/evalmapbench/): as of 2026-11-06; licence Private data · only these aggregates are published (https://evalmap.ai/licenses/)
- Epoch AI (https://epoch.ai/benchmarks): as of 2026-10-18; licence CC BY 4.0 (https://epoch.ai/benchmarks/use-this-data)
- LMArena leaderboard (https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset): as of 2026-10-08 to 2026-10-14; licence CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/)
- models.dev catalogue (https://models.dev/): as of 2026-10-18; licence MIT (https://github.com/anomalyco/models.dev/blob/dev/LICENSE)
- LiteLLM model prices (https://github.com/BerriAI/litellm/blob/main/model_prices_and_context_window.json): as of 2026-10-18; licence MIT (https://github.com/BerriAI/litellm/blob/main/LICENSE)
- Vals AI (https://www.vals.ai/benchmarks): as of 2026-09-01 to 2026-10-16; licence No licence and no terms published · republished with attribution (https://www.vals.ai/methodology)
- BridgeBench (https://www.bridgebench.ai/leaderboard): as of 2026-10-15; licence No licence and no terms published · republished with attribution (https://www.bridgebench.ai/blog/how-the-leaderboard-works)
- EvalMap measurement (https://evalmap.ai/licenses/): as of 2026-09-22; licence Measured here; every run is kept with the site’s sources (https://evalmap.ai/licenses/)

## Questions

### What is EvalMap?

Compare 88 AI model configurations in 3D: coding, reasoning and arena benchmarks against speed and price per million tokens. Every value names its source. Values are never blended, and nothing is ranked or recommended.

### Where do the numbers come from?

From Artificial Analysis, Terminal-Bench 4.0 official board, DeepSWE v1.1, EvalMapBench 6.7, Epoch AI, LMArena leaderboard, models.dev catalogue, LiteLLM model prices, Vals AI, BridgeBench, EvalMap measurement. Every value names its source; each source and its licence is listed at https://evalmap.ai/licenses/.

### How current is the data?

Each source is a dated snapshot: Artificial Analysis 4 November 2026; Terminal-Bench 4.0 official board 2 September 2026; DeepSWE v1.1 22 September 2026; EvalMapBench 6.7 6 November 2026; Epoch AI 18 October 2026; LMArena leaderboard 2026-10-08 to 2026-10-14; models.dev catalogue 18 October 2026; LiteLLM model prices 18 October 2026; Vals AI 2026-09-01 to 2026-10-16; BridgeBench 15 October 2026; EvalMap measurement 22 September 2026. Nothing on the site updates by itself.

### Which models score highest on Terminal-Bench 4.0?

In the Artificial Analysis snapshot: Kimi K3.1 Pro 67.2%; Gemini 4 Carbon (High) 65.2%; Claude Sonnet 5.5 (max) 63.6% (Terminal-Bench as Artificial Analysis runs it, not the official leaderboard).

### What is EvalMapBench?

47 AI models on 43 private maintenance tickets from one developer’s own services, each written tersely and checked by a hidden verifier for every detail the ticket left unsaid, scored beyond a gpt-oss-120b (high) baseline, against what each costs per million output tokens.

### What is the jeep gallery?

The same brief, a drivable 3D jeep race in a browser page, given to 142 AI models and coding agents in 2026. Every build that runs plays here, scored 0–100.