OSWorld-Verified leaderboard

OSWorld-Verified is the July 2025 repaired release of OSWorld's real-computer evaluation. It measures whether a model-agent system can finish desktop and web tasks from configured starting states, with success checked by execution-based evaluators.

Models ranked by their published OSWorld-Verified value

Rank	Model	Provider	Execution-based interactive task success
1	Qwen3.8 Max	Alibaba	86.1
2	Claude Fable 5	Anthropic	85
2	Claude Mythos 5	Anthropic	85
4	Qwen3.8-27B	Alibaba	84.3
5	Claude Opus 4.8	Anthropic	83.4
6	Gemini 3.6 Flash	Google	83
7	HCHolo3-35B-A3B	H Company	82.56
8	Claude Sonnet 5	Anthropic	81.2
9	Muse Spark 1.1	Meta	80.8
10	HCHolo3-122B-A10B	H Company	78.85
11	GPT-5.5	OpenAI	78.7
12	Gemini 3.5 Flash	Google	78.4
13	Claude Opus 4.7 (Adaptive)	Anthropic	78
14	UI-Mate-27B	Tencent	77
15	GPT-5.4	OpenAI	75
16	Gemini 3.5 Flash-Lite	Google	74
17	Qwen3.7 Plus	Alibaba	73.3
18	Kimi K2.6	Moonshot AI	73.1
19	Claude Opus 4.6	Anthropic	72.7
20	Claude Sonnet 4.6	Anthropic	72.1
20	GPT-5.4 mini	OpenAI	72.1
22	MiniMax M3	MiniMax	70.06
23	Claude Opus 4.5	Anthropic	66.3
24	UI-Mate-9B	Tencent	66.2
25	Muse Glimmer 30B	Meta	65.9
26	GPT-5.3 Codex	OpenAI	64.7
27	Claude Sonnet 4.5	Anthropic	61.4
28	Qwen3.5-122B-A10B	Alibaba	58
29	Qwen3.5-27B	Alibaba	56.2
30	Qwen3.5-35B-A3B	Alibaba	54.5
31	GPT-5.2	OpenAI	47.3
32	GPT-5.4 nano	OpenAI	39

Evidence key: Observed

Rows are ordered by the value OSWorld published, highest first. The source does not state whether a higher value is the better result, so this page does not either: for a benchmark that measures a rate of failure, read the table from the bottom.