Surface Evolver Bench
ReportedNo description yet: we only describe an eval set when we can do it correctly.
- Kind
- Published by a third party's leaderboard, collected by Epoch AI. The leaderboard is named and linked.
- Link
- https://yhenon.github.io/surface-evolver-llm-eval/
- Score shown
- Epoch's column "Mean score", as a percentage, higher is better; random guessing scores about 0.0%. The source gives no error bars.
- Results
- 29 in all: 16 for open-weight models, 13 for closed ones. 13 linked to a model page here.
Best open-weight result: moonshotai/Kimi-K3, 95.0%, number 2 of 29 results.
| # | Model | Weights | Setting | Score | Date | Source |
|---|---|---|---|---|---|---|
| 1 | claude-fable-5_highAnthropic | closed | effort high | 95.0% | not dated | yhenon.github.io |
| 2 | moonshotai/Kimi-K3Epoch: kimi-k3_unknown, Moonshot | open | effort unknown | 95.0% | not dated | yhenon.github.io |
| 3 | gpt-5.6-sol_xhighOpenAI | closed | effort xhigh | 93.1% | not dated | yhenon.github.io |
| 4 | moonshotai/Kimi-K3Epoch: kimi-k3_max, Moonshot | open | effort max | 93.0% | not dated | yhenon.github.io |
| 5 | gpt-5.5_highOpenAI | closed | effort high | 88.1% | not dated | yhenon.github.io |
| 6 | claude-opus-4-8_highAnthropic | closed | effort high | 87.5% | not dated | yhenon.github.io |
| 7 | gpt-5.6-terra_xhighOpenAI | closed | effort xhigh | 83.8% | not dated | yhenon.github.io |
| 8 | gpt-5.5_mediumOpenAI | closed | effort medium | 81.3% | not dated | yhenon.github.io |
| 9 | gemini-3.8-flash_highGoogle DeepMind | closed | effort high | 76.9% | not dated | yhenon.github.io |
| 10 | grok-4.5_highxAI | closed | effort high | 74.4% | not dated | yhenon.github.io |
| 11 | claude-opus-4-8_noneAnthropic | closed | effort none | 68.1% | not dated | yhenon.github.io |
| 12 | gpt-5.6-luna_mediumOpenAI | closed | effort medium | 61.9% | not dated | yhenon.github.io |
| 13 | claude-sonnet-5_mediumAnthropic | closed | effort medium | 60.0% | not dated | yhenon.github.io |
| 14 | gemini-3.5-flash_mediumGoogle DeepMind | closed | effort medium | 58.1% | not dated | yhenon.github.io |
| 15 | zai-org/GLM-5.2Epoch: glm-5.2_high, Z.ai (Zhipu AI) | open | effort high | 55.6% | not dated | yhenon.github.io |
| 16 | MiniMaxAI/MiniMax-M3Epoch: MiniMax-M3, MiniMax | open | default | 55.0% | not dated | yhenon.github.io |
| 17 | zai-org/GLM-5.3-FlashEpoch: glm-5.3-flash_max, Z.ai (Zhipu AI) | open | effort max | 52.5% | not dated | yhenon.github.io |
| 18 | muse-spark-1.1_highMeta AI | closed | effort high | 52.5% | not dated | yhenon.github.io |
| 19 | moonshotai/Kimi-K2.7-CodeEpoch: kimi-k2.7-code, Moonshot | open | default | 48.8% | not dated | yhenon.github.io |
| 20 | deepseek-ai/DeepSeek-V4.1-FlashEpoch: deepseek-v4.1-flash_high, DeepSeek | open | effort high | 46.3% | not dated | yhenon.github.io |
| 21 | Qwen/Qwen3.8-27BEpoch: qwen3.8-27b_xhigh, Alibaba | open | effort xhigh | 45.0% | not dated | yhenon.github.io |
| 22 | Qwen/Qwen3.6-35B-A3BEpoch: qwen3.6-35b-a3b, Alibaba | open | default | 44.4% | not dated | yhenon.github.io |
| 23 | deepseek-ai/DeepSeek-V4-ProEpoch: deepseek-v4-pro_high, DeepSeek | open | effort high | 40.0% | not dated | yhenon.github.io |
| 24 | google/gemma-4-31B-itEpoch: gemma-4-31b-it, Google DeepMind | open | default | 30.6% | not dated | yhenon.github.io |
| 25 | mistral-large-4-0_unknownMistral AIno page match | open | effort unknown | 28.1% | not dated | yhenon.github.io |
| 26 | mistral-medium-2604Mistral AIno page match | open | default | 26.9% | not dated | yhenon.github.io |
| 27 | openai/gpt-oss-120bEpoch: gpt-oss-120b, OpenAI | open | default | 25.0% | not dated | yhenon.github.io |
| 28 | poolside/Laguna-M.1Epoch: laguna-m.1, Poolside | open | default | 15.6% | not dated | yhenon.github.io |
| 29 | trinity-large-thinkingArcee AIno page match | open | default | 15.6% | not dated | yhenon.github.io |
Open-weight rows are shaded. Dates are when the result was run or evaluated, where the data says. Data from Epoch AI (CC-BY 4.0), fetched 19 Nov 2026: Epoch AI, ‘Capabilities & benchmarking’. Published online at epoch.ai. Retrieved from ‘https://epoch.ai/benchmarks’ [online resource]. Epoch AI. Method