OSWorld-Verified 2.0

Computer-use tasks in a real desktop environment.

Published by

OSWorlddata via Epoch AI Benchmarking Hub

Category

Agents & tools

Index weight

0.5

Models

14

Data as of

9 Sept 2026

CC BY 4.0 — Epoch AI, ‘AI Benchmarking Hub’, epoch.ai/benchmarks. Mirrored boards credit their original publishers.

1. 1Claude Opus 531.4%
2. 2Claude Opus 530.2%
3. 3Claude Opus 529.0%
4. 4GPT-5.6 Sol27.3%
5. 5Claude Opus 525.3%
6. 6Claude Opus 522.3%
7. 7Claude Opus 4.820.6%
8. 8Claude Opus 4.718.2%
9. 9GPT-5.513.0%
10. 10Claude Sonnet 4.69.3%
11. 11Claude Sonnet 4.68.3%
12. 12Kimi K2.64.6%
13. 13MiniMax-M34.6%
14. 14Qwen3.7 Plus2.8%

SearchAll providersAlibabaAnthropicMiniMaxMoonshot AIOpenAIOpen weights14 of 14

#	Model	Score ↓	Run ↕	Index ↕	Published ↕
1	Claude Opus 5Anthropic	31.4%	–	66.8	–
2	Claude Opus 5Anthropic	30.2%	–	66.3	–
3	Claude Opus 5Anthropic	29.0%	–	69.3	–
4	GPT-5.6 SolOpenAI	27.3%	–	65.5	–
5	Claude Opus 5Anthropic	25.3%	–	63.1	–
6	Claude Opus 5Anthropic	22.3%	–	61.7	–
7	Claude Opus 4.8Anthropic	20.6%	–	59.4	–
8	Claude Opus 4.7Anthropic	18.2%	–	59.3	–
9	GPT-5.5OpenAI	13.0%	–	63.7	–
10	Claude Sonnet 4.6Anthropic	9.3%	–	51.8	–
11	Claude Sonnet 4.6Anthropic	8.3%	–	49.7	–
12	Kimi K2.6Moonshot AIopen	4.6%	Kimi 2.6 (enabled)	60.6	–
13	MiniMax-M3MiniMaxopen	4.6%	MiniMax M3 (enabled)	56.9	–
14	Qwen3.7 PlusAlibaba	2.8%	–	59.7	–