Surface Evolver Bench leaderboard
- Claude Fable 5 leads
- 95.0%
- models measured
- 25
- at the top
- Tied
- median
- 55.0%
Leader on Surface Evolver Bench · read October 22
On October 22, 2026, 25 of our catalog's models had a Surface Evolver Bench result, and Anthropic Claude Fable 5 led with 95.0%.
Epoch AI collects its results; its benchmarking hub describes the tasks and links the authors.
Role in the index
Surface Evolver Bench is shown as published and does not feed any of our indexes. Read the methodology
| # | Model | Result | Cost of the run | Pass rate | Tokens for the run | Evidence | Coding index | Output $/Mtok |
|---|---|---|---|---|---|---|---|---|
| 1 | Claude Fable 5Anthropic | 95.0% | $19.4352 | 87.5% | 933,342 | 79.3 | $50.00 | |
| 2 | Kimi K3Moonshot AI | 95.0% | $12.1422 | 87.5% | 3.7M | 62.3 | $15.00 | |
| 3 | GPT-5.6 SolOpenAI | 93.1% | $9.4078 | 87.5% | 1.4M | 73.5 | $20.00 | |
| 4 | GPT-5.5OpenAI | 88.1% | $13.8258 | 75.0% | 2.0M | 71.3 | $30.00 | |
| 5 | Claude Opus 4.8Anthropic | 87.5% | $21.9616 | 68.8% | 2.7M | 65.0 | $25.00 | |
| 6 | GPT-5.6 TerraOpenAI | 83.8% | $7.5899 | 68.8% | 2.1M | 63.9 | $12.00 | |
| 7 | Gemini 3.8 FlashGoogle | 76.9% | $2.5981 | 56.3% | 3.2M | 61.6 | $3.75 | |
| 8 | Grok 4.5xAI | 74.4% | $4.6103 | 50.0% | 3.1M | 58.6 | $6.00 | |
| 9 | GPT-5.6 LunaOpenAI | 61.9% | $1.0079 | 18.8% | 1.2M | 58.0 | $1.20 | |
| 10 | Claude Sonnet 5Anthropic | 60.0% | $11.9389 | 18.8% | 3.8M | 57.7 | $10.00 | |
| 11 | Gemini 3.5 FlashGoogle | 58.1% | $6.6382 | 37.5% | 3.1M | 39.8 | $9.00 | |
| 12 | GLM 5.2Z.ai | 55.6% | $4.6462 | 31.3% | 2.5M | 61.1 | $4.40 | |
| 13 | MiniMax M3MiniMax | 55.0% | $2.0781 | 18.8% | 5.5M | 39.2 | $1.20 | |
| 14 | Muse Spark 1.1Meta | 52.5% | $2.4983 | 37.5% | 3.8M | 70.3 | $4.25 | |
| 15 | GLM 5.3 FlashZ.ai | 52.5% | $0.3085 | 18.8% | 2.6M | 57.7 | $0.50 | |
| 16 | Kimi K2.7 CodeMoonshot AI | 48.8% | $4.0321 | 37.5% | 5.3M | 53.4 | $4.00 | |
| 17 | DeepSeek V4.1 FlashDeepSeek | 46.3% | $2.1277 | 31.3% | 3.3M | 62.5 | $0.84 | |
| 18 | Qwen3.8 27BQwen | 45.0% | $9.0488 | 18.8% | 4.7M | 74.2 | $2.55 | |
| 19 | Qwen3.6 35B A3BQwen | 44.4% | $2.1926 | 12.5% | 4.0M | Not benchmarked | $1.00 | |
| 20 | DeepSeek V4 Pro 0423DeepSeek | 40.0% | $3.857 | 25.0% | 4.5M | 38.7 | $2.976 | |
| 21 | Gemma 4 31BGoogle | 30.6% | $0.2156 | 6.3% | 1.4M | 39.0 | $0.40 | |
| 22 | Mistral Large 4Mistral AI | 28.1% | $6.0172 | 18.8% | 5.7M | 56.4 | $4.18 | |
| 23 | Mistral Medium 3.5Mistral AI | 26.9% | $6.3482 | 6.3% | 2.8M | 25.5 | $7.50 | |
| 24 | gpt-oss-120bOpenAI | 25.0% | $0.0892 | 6.3% | 1.5M | Not benchmarked | $0.50 | |
| 25 | Trinity Large ThinkingArcee AI | 15.6% | $1.2472 | 6.3% | 2.6M | Not benchmarked | $0.80 |
Only models in our catalog are listed, and each appears once with its best published run. 215 rows were read from the source on October 22, 2026, and rows that match no catalog model are left out. Cost of the run is what the whole run cost, in US dollars, as Epoch AI records it beside the result. Every other column is a figure the board publishes for the same run.