Surface evolver bench — Epoch AI

Surface evolver bench — Epoch AI is run by Epoch AI. It has ranked 26 models, of which the catalogue holds 22, scoring from 0.156 to 0.95.

measured by Epoch AI · the board itself

Place	Model	Metric	Score
1stof 26	Claude Fable 5	Mean score (high)	0.95
2ndof 26	Kimi K3	Mean score	0.95
3rdof 26	GPT-5.6 Sol	Mean score (xhigh)	0.931
5thof 26	GPT-5.5	Mean score (high)	0.881
6thof 26	Claude Opus 4.8	Mean score (high)	0.875
7thof 26	GPT-5.6 Terra	Mean score (xhigh)	0.838
9thof 26	Grok 4.5	Mean score (high)	0.744
11thof 26	GPT-5.6 Luna	Mean score (medium)	0.619
12thof 26	Claude Sonnet 5	Mean score (medium)	0.6
13thof 26	Gemini 3.5 Flash	Mean score (medium)	0.581
14thof 26	GLM 5.2	Mean score (high)	0.556
15thof 26	MiniMax M3	Mean score	0.55
16thof 26	Muse Spark 1.1	Mean score (high)	0.525
17thof 26	GLM 5.3 Flash	Mean score (max)	0.525
18thof 26	Kimi K2.7 Code	Mean score	0.487
19thof 26	Qwen3.8 27B	Mean score (xhigh)	0.45
20thof 26	Qwen3.6 35B A3B	Mean score	0.444
21stof 26	DeepSeek V4 Pro	Mean score (high)	0.4
22ndof 26	Gemma 4 31B	Mean score	0.306
24thof 26	GPT OSS 120B	Mean score	0.25
25thof 26	Laguna M.1	Mean score	0.156
26thof 26	Trinity Large Thinking	Mean score	0.156

Place

Model

Metric

Score

1stof 26

Mean score (high)

0.95

2ndof 26

Mean score

0.95