Surface Evolver Bench
- What does it measure?
- Tests whether an AI agent can use Surface Evolver, a program that models liquid surfaces shaped by surface tension.
- Who evaluates it, and how?
- The agent looks things up in the documentation and writes a complete input file meeting the given conditions (geometry, contact angles, gravity, volume). It is scored as the average partial credit across tasks.
- Reading the score
- Scores run from 0 to 100%. Higher is better.
Top score95.0%
Models tested25
Scoring basisHigher is better
Model release date
Can it solve problems that stump experts across fields?Humanity's Last Exam
Claude Opus 5.5🥇
Claude Opus 5.5🥇Humanity's Last Exam
Claude Opus 5.5🥇Can it solve research-level physics problems?CritPt
Gemini 4 Carbon🥇
Gemini 4 Carbon🥇CritPt
Gemini 4 Carbon🥇Which AI gives the math answers people prefer most?Arena Math
Gemini 4 Argon🥇
Gemini 4 Argon🥇Arena Math
Gemini 4 Argon🥇
Can it run scientific simulation software to spec?
Surface Evolver Bench
- 🥇AnthropicClaude Fable 5Reasoning effort: High95.0%
- 🥇Moonshot AIKimi K395.0%
- 🥉OpenAIGPT-5.6 SolReasoning effort: Extra High93.1%
- #4OpenAIGPT-5.5Reasoning effort: High88.1%
- #5AnthropicClaude Opus 4.8Reasoning effort: High87.5%
- #6OpenAIGPT-5.6 TerraReasoning effort: Extra High83.8%
- #7GoogleGemini 3.8 FlashReasoning effort: High76.9%
- #8SpaceXSIGrok 4.5Reasoning effort: High74.4%
- #9OpenAIGPT-5.6 LunaReasoning effort: Medium61.9%
- #10AnthropicClaude Sonnet 5Reasoning effort: Medium60.0%
- #11GoogleGemini 3.5 FlashReasoning effort: Medium58.1%
- #12Z.aiGLM 5.2Reasoning effort: High55.6%
- #13MiniMaxMiniMax M355.0%
- #14MetaMuse Spark 1.1Reasoning effort: High52.5%
- #14Z.aiGLM 5.3 FlashReasoning effort: Max52.5%
- #16Moonshot AIKimi K2.7 Code48.8%
- #17DeepSeekDeepSeek V4.1 FlashReasoning effort: High46.3%
- #18AlibabaQwen3.6 35B A3B44.4%
- #19DeepSeekDeepSeek V4 ProReasoning effort: High40.0%
- #20GoogleGemma 4 31B30.6%
- #21Mistral AIMistral Large 428.1%
- #22Mistral AIMistral Medium 3.526.9%
- #23OpenAIGPT OSS 120B25.0%
- #24PoolsideLaguna M.115.6%
- #24Arcee AITrinity Large Thinking15.6%
Half of models ≤ 55.0%
The "harness" is the agent program the AI used to carry out the task. The same model can score very differently depending on its harness and reasoning effort.