AA Terminal-Bench Science 0.1 - Physical Sciences: leaderboard
Artificial Analysis pass@1 on the 17 physical science tasks in Terminal-Bench Science 0.1.0. The mini-swe-agent harness uses a 1,000-step cap, upstream resource/time limits and isolated verifier containers. All tests must pass; scores average three repeats per task and are percentages.
Metric: Pass@1 (%). Source: artificialanalysis.ai. Saturation forecast: Around May 2027. 100 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-6.1 Sol (Max) | 66.67 | #7 (GPT-6.1 Sol) |
| 2 | GPT-6.1 Terra (Max) | 60.78 | #13 (GPT-6.1 Terra) |
| 3 | GPT-6 Astra (xHigh) | 56.86 | #5 (GPT-6 Astra) |
| 4 | GPT-6 Astra (Medium) | 56.86 | #5 (GPT-6 Astra) |
| 5 | GPT-6 Astra (Max) | 54.9 | #5 (GPT-6 Astra) |
| 6 | GPT-6 Astra (High) | 54.9 | #5 (GPT-6 Astra) |
| 7 | Gemini 4 Carbon (High) | 54.9 | #2 (Gemini 4 Carbon) |
| 8 | GPT-6.1 Sol (xHigh) | 50.98 | #7 (GPT-6.1 Sol) |
| 9 | GPT-6.1 Sol (High) | 49.02 | #7 (GPT-6.1 Sol) |
| 10 | GPT-6.1 Terra (xHigh) | 49.02 | #13 (GPT-6.1 Terra) |
| 11 | Gemini 4 Argon (High) | 47.06 | #9 (Gemini 4 Argon) |
| 12 | GPT-6.1 Terra (High) | 41.18 | #13 (GPT-6.1 Terra) |
| 13 | Kimi K3.1 Pro | 39.22 | #16 |
| 14 | GPT-6 Astra (Low) | 39.22 | #5 (GPT-6 Astra) |
| 15 | GPT-6.1 Sol (Medium) | 37.25 | #7 (GPT-6.1 Sol) |
No result here: #3 Claude Opus 5.5, #10 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=aa-terminal-bench-science-0-1-physical-sciences · How It Works · Data refreshed daily, snapshot 2026-10-29.