AA Coding Agent Index v2.0 - Terminal-Bench 4.0: leaderboard
Terminal-Bench 4.0 component of AA's Coding Agent Index v2.0: 66 agentic terminal tasks solved by each model inside the coding-agent harness named in its row, three attempts per task, test-suite pass or fail. Separate from the model-level AA Terminal-Bench 4.0 board, which runs every model in one harness. Pass@1 (%).
Metric: Pass@1 (%). Source: artificialanalysis.ai. Saturation forecast: Around October 2027. 30 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 4 Carbon (High) | 63.64 | #1 (Gemini 4 Carbon) |
| 2 | Muse Wonder Full (Max) | 59.09 | #12 (Muse Wonder Full) |
| 3 | Gemini 4 Argon (High) | 56.06 | #9 (Gemini 4 Argon) |
| 4 | GPT-6 Astra (Max) | 55.56 | #6 (GPT-6 Astra) |
| 5 | GPT-6.1 Sol (Max) | 53.03 | #7 (GPT-6.1 Sol) |
| 6 | GPT-6.1 Terra (Max) | 48.48 | #15 (GPT-6.1 Terra) |
| 7 | Muse Wonder (Max) | 47.98 | #18 (Muse Wonder) |
| 8 | GPT-6.1 Terra (xHigh) | 45.96 | #15 (GPT-6.1 Terra) |
| 9 | GPT-6.1 Terra (High) | 42.42 | #15 (GPT-6.1 Terra) |
| 10 | mistral-large-4-oss (High) | 41.41 | #60 (mistral-large-4-oss) |
| 11 | GLM-5.3 (Max) | 39.9 | #44 (GLM-5.3) |
| 12 | GPT-6.1 Terra (Medium) | 35.35 | #15 (GPT-6.1 Terra) |
| 13 | Grok 4.7 (xHigh) | 33.33 | #26 (Grok 4.7) |
| 14 | Muse Spark 1.3 (Max) | 31.82 | #31 (Muse Spark 1.3) |
| 15 | DeepSeek V4.1 Flash (1017) (Max) | 30.81 | #27 (DeepSeek V4.1 Flash (1017)) |
No result here: #3 Claude Opus 5.5, #10 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=aa-coding-agent-index-v2-0-terminal-bench-4-0 · How It Works · Data refreshed daily, snapshot 2026-11-05.