Surface Evolver Bench: leaderboard
Surface Evolver Bench evaluates agentic scientific simulation writing, with models creating liquid-surface physics datafiles and using tool feedback before hidden checks.
Metric: Mean Score (%). Source: yhenon.github.io. Saturation forecast: Around January 2027. 28 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Kimi K3 | 95 | #30 |
| 2 | Claude Fable 5 (High) | 95 | #16 (Claude Fable 5) |
| 3 | GPT-5.6 Sol (xHigh) | 93.12 | #19 (GPT-5.6 Sol) |
| 4 | GPT-5.5 (High) | 88.12 | #31 (GPT-5.5) |
| 5 | Claude Opus 4.8 (High) | 87.5 | #37 (Claude Opus 4.8) |
| 6 | GPT-5.6 Terra (xHigh) | 83.75 | #41 (GPT-5.6 Terra) |
| 7 | GPT-5.5 (Medium) | 81.25 | #31 (GPT-5.5) |
| 8 | Gemini 3.8 Flash (High) | 76.88 | #21 (Gemini 3.8 Flash) |
| 9 | Grok 4.5 (High) | 74.38 | #48 (Grok 4.5) |
| 10 | Claude Opus 4.8 | 68.12 | #37 |
| 11 | GPT-5.6 Luna (Medium) | 61.88 | #68 (GPT-5.6 Luna) |
| 12 | Claude Sonnet 5 (Medium) | 60 | #57 (Claude Sonnet 5) |
| 13 | Gemini 3.5 Flash (Medium) | 58.13 | #60 (Gemini 3.5 Flash) |
| 14 | GLM-5.2 (High) | 55.62 | #73 (GLM-5.2) |
| 15 | MiniMax-M3 | 55 | #136 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=surface-evolver-bench · How It Works · Data refreshed daily, snapshot 2026-10-20.