Osworld 2 — Epoch AI

Osworld 2 — Epoch AI is run by Epoch AI. It has ranked 16 models, of which the catalogue holds 9, scoring from 0.028 to 0.314.

measured by Epoch AI · the board itself

Place	Model	Metric	Score
1stof 16	Claude Opus 5	Binary accuracy (max)	0.314
4thof 16	GPT-5.6 Sol	Binary accuracy (max)	0.273
7thof 16	Claude Opus 4.8	Binary accuracy (max)	0.206
9thof 16	Claude Opus 4.7	Binary accuracy (max)	0.182
11thof 16	GPT-5.5	Binary accuracy (xhigh)	0.13
12thof 16	Claude Sonnet 4.6	Binary accuracy (medium)	0.093
14thof 16	MiniMax M3	Binary accuracy	0.046
15thof 16	Kimi K2.6	Binary accuracy	0.046
16thof 16	Qwen 3.7 Plus	Binary accuracy	0.028

Place

Model

Metric

Score

1stof 16

Binary accuracy (max)

0.314

4thof 16

Binary accuracy (max)

0.273