Agentic · tools

OSWorld 2.0 leaderboard

Name: benchgap: LLM benchmark scores, measured and estimated
Creator: benchgap

As of 2026-10-07, the highest measured score on OSWorld 2.0 is 72.6% by GPT-6 Astra. 99 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	GPT-6 Astra	72.6%	measured
2	Claude Opus 5	70.6%	measured
3	Gemini 4 Argon	69.2%	measured
4	Atria Dawn Preview	68.6%	estimated ± 13.8 pp, low confidence
5	GPT-5.5 Pro	68.5%	estimated ± 13.8 pp, low confidence
6	GPT-5.4 Pro	67.8%	estimated ± 13.8 pp, low confidence
7	Muse Spark 1.3	66.9%	measured
8	GPT-5.6 Sol	62.6%	measured
9	DeepSeek V4.1 Flash	61.0%	estimated ± 13.3 pp, low confidence
10	GPT-6 Sol	60.5%	measured
11	Grok 4.6	59.6%	estimated ± 13.3 pp, low confidence
12	Gemini 3.8 Flash	59.0%	measured
13	GLM-5.3-Flash	59.0%	estimated ± 5.5 pp, low confidence
14	Grok 4.7	58.9%	estimated ± 13.3 pp, low confidence
15	GPT-6.1 Sol	58.5%	estimated ± 13.3 pp, low confidence
16	Mistral Large 4	55.2%	estimated ± 13.3 pp, low confidence
17	GPT-6 Luna	50.6%	estimated ± 13.3 pp, low confidence
18	GPT-5.6 Terra	50.2%	measured
19	Claude Opus 5.5	48.7%	measured
20	Gemini 3.7 Flash	47.9%	measured
21	GPT-5.6 Luna	45.6%	measured
22	Claude Sonnet 5.5	45.2%	estimated ± 5.5 pp, low confidence
23	Claude Fable 5.1	41.7%	measured
24	Claude Mythos Preview	40.2%	estimated ± 13.6 pp, low confidence
25	Claude Haiku 5.5	36.8%	estimated ± 13.3 pp, low confidence
26	Claude Opus 4.8	20.6%	measured
27	Claude Fable 5	20.4%	estimated ± 3.6 pp, high confidence
28	Claude Mythos 5	20.4%	estimated ± 3.6 pp, high confidence
29	Qwen3.8-27B	19.6%	estimated ± 3.6 pp, high confidence
30	DeepSeek V4 Pro 0813	19.4%	estimated ± 5.5 pp, low confidence
31	Hy4 preview	19.4%	estimated ± 5.5 pp, low confidence
32	Step 5 Preview	19.4%	estimated ± 5.5 pp, low confidence
33	Qwen3.8-Flash-Next	19.4%	measured
34	Qwen3.8 Max	19.4%	measured
35	Kimi K3	19.4%	estimated ± 5.5 pp, low confidence
36	GLM-5.3	19.4%	estimated ± 5.5 pp, low confidence
37	Ornith-1.5-397B	19.4%	estimated ± 5.5 pp, low confidence
38	DeepSeek V4 Flash 0731	19.4%	estimated ± 5.5 pp, low confidence
39	dots3-note Preview	19.4%	estimated ± 5.5 pp, low confidence
40	Inkling-Small	19.4%	estimated ± 5.5 pp, low confidence
41	Laguna S 2.1	19.4%	estimated ± 5.5 pp, low confidence
42	Ornith-1.5-35B-A3B	19.4%	estimated ± 5.5 pp, low confidence
43	Ornith-1.5-9B	19.4%	estimated ± 5.5 pp, low confidence
44	Claude Opus 4.7 (Adaptive)	18.2%	measured
45	Gemini 3.6 Flash	18.1%	estimated ± 3.6 pp, high confidence
46	Agents-A1	17.7%	estimated ± 13.8 pp, low confidence
47	Agents-A1-4B	17.7%	estimated ± 13.8 pp, low confidence
48	Ling 3.0 Flash	17.7%	estimated ± 13.8 pp, low confidence
49	LongCat-Flash-Lite-Sparse	17.7%	estimated ± 13.8 pp, low confidence
50	Nemotron 3.5 Lightning 30B A3B NVFP4	17.7%	estimated ± 13.8 pp, low confidence
51	Beam	17.7%	estimated ± 13.8 pp, low confidence
52	Solar Pro 4	17.7%	estimated ± 13.8 pp, low confidence
53	Holo3-35B-A3B	17.7%	estimated ± 3.6 pp, high confidence
54	MiMo-V2.6-Pro	17.0%	estimated ± 3.6 pp, high confidence
55	Claude Sonnet 5	16.1%	estimated ± 3.6 pp, high confidence
56	MiMo-V2.6-Flash	15.7%	estimated ± 3.6 pp, high confidence
57	Muse Spark 1.1	14.2%	measured
58	Claude Opus 4.7	13.9%	measured
59	GLM-5.2	13.6%	estimated ± 5.1 pp, low confidence
60	Holo3-122B-A10B	13.5%	estimated ± 3.6 pp, high confidence
61	GPT-5.5	13.0%	measured
62	Gemini 3.5 Flash	13.0%	estimated ± 3.6 pp, high confidence
63	UI-Mate-27B	11.4%	estimated ± 3.6 pp, high confidence
64	Qwen3.7 Max	10.6%	estimated ± 4.5 pp, medium confidence
65	Gemini 3 Pro	9.8%	estimated ± 4.5 pp, medium confidence
66	MiMo-V2.5-Pro	9.4%	estimated ± 4.5 pp, medium confidence
67	GPT-5.4	9.2%	estimated ± 3.6 pp, high confidence
68	Claude Sonnet 4.6	8.3%	measured
69	Gemini 3.5 Flash-Lite	8.1%	estimated ± 3.6 pp, high confidence
70	GLM-5.1	7.4%	estimated ± 4.5 pp, medium confidence
71	Grok 4.1	7.2%	estimated ± 5.1 pp, low confidence
72	Claude Opus 4.6	6.6%	estimated ± 3.6 pp, high confidence
73	Inkling	6.2%	estimated ± 13.3 pp, low confidence
74	GPT-5.4 mini	5.9%	estimated ± 3.6 pp, high confidence
75	Gemini 3.1 Pro	4.9%	estimated ± 4.5 pp, medium confidence
76	Gemini 3 Flash	4.7%	estimated ± 4.5 pp, low confidence
77	Kimi K2.6	4.6%	measured
78	MiniMax M3	4.6%	measured
79	Nemotron 3 Ultra	3.7%	estimated ± 13.3 pp, low confidence
80	Qwen3.6-27B	3.4%	estimated ± 4.5 pp, low confidence
81	Qwen3.7 Plus	2.8%	measured
82	GPT-5.2-Codex	1.0%	estimated ± 4.5 pp, low confidence
83	Step 3.7 Flash	0.9%	estimated ± 4.5 pp, low confidence
84	GLM-5	0.4%	estimated ± 4.5 pp, low confidence
85	Qwen3.6 Plus	0.1%	estimated ± 4.5 pp, low confidence
86	Claude 4 Sonnet	0.0%	estimated ± 4.5 pp, low confidence
87	Claude Opus 4.5	0.0%	estimated ± 3.6 pp, medium confidence
88	Claude Sonnet 4.5	0.0%	estimated ± 3.6 pp, medium confidence
89	DeepSeek V3.2	0.0%	estimated ± 4.5 pp, low confidence
90	Gemini 2.5 Pro	0.0%	estimated ± 4.5 pp, low confidence
91	Gemini 3.1 Flash-Lite	0.0%	estimated ± 4.5 pp, low confidence
92	Gemma 4 31B	0.0%	estimated ± 4.5 pp, low confidence
93	GLM-4.7	0.0%	estimated ± 4.5 pp, low confidence
94	GLM-5V-Turbo	0.0%	estimated ± 4.5 pp, low confidence
95	GPT-4.1	0.0%	estimated ± 4.5 pp, low confidence
96	GPT-5.1	0.0%	estimated ± 4.5 pp, low confidence
97	GPT-5.1-Codex	0.0%	estimated ± 4.5 pp, low confidence
98	GPT-5.2	0.0%	estimated ± 3.6 pp, medium confidence
99	GPT-5.3 Codex	0.0%	estimated ± 3.6 pp, medium confidence
100	GPT-5.4 nano	0.0%	estimated ± 3.6 pp, medium confidence
101	GPT-OSS 120B	0.0%	estimated ± 4.5 pp, low confidence
102	Grok 4	0.0%	estimated ± 4.5 pp, low confidence
103	Grok 4.1 Fast	0.0%	estimated ± 4.5 pp, low confidence
104	Grok 4.20	0.0%	estimated ± 4.5 pp, low confidence
105	Grok 4.3	0.0%	estimated ± 4.5 pp, low confidence
106	Grok Build 0.1	0.0%	estimated ± 4.5 pp, low confidence
107	Hy3 Preview	0.0%	estimated ± 4.5 pp, low confidence
108	Kimi K2.5	0.0%	estimated ± 4.5 pp, low confidence
109	Kimi K2.5 (Reasoning)	0.0%	estimated ± 4.5 pp, low confidence
110	MiMo-V2.5	0.0%	estimated ± 4.5 pp, low confidence
111	MiMo-V2-Pro	0.0%	estimated ± 4.5 pp, low confidence
112	MiniMax M2.7	0.0%	estimated ± 4.5 pp, low confidence
113	Mistral Medium 3.5 128B	0.0%	estimated ± 4.5 pp, low confidence
114	Muse Glimmer 30B	0.0%	estimated ± 3.6 pp, medium confidence
115	Qwen3.5-122B-A10B	0.0%	estimated ± 3.6 pp, medium confidence
116	Qwen3.5-27B	0.0%	estimated ± 3.6 pp, medium confidence
117	Qwen3.5-35B-A3B	0.0%	estimated ± 3.6 pp, medium confidence
118	Qwen3.5 397B	0.0%	estimated ± 4.5 pp, low confidence
119	Qwen3.6-35B-A3B	0.0%	estimated ± 4.5 pp, low confidence
120	Qwen3 Max	0.0%	estimated ± 4.5 pp, low confidence
121	Trinity-Large-Thinking	0.0%	estimated ± 4.5 pp, low confidence
122	UI-Mate-9B	0.0%	estimated ± 3.6 pp, medium confidence

OSWorld 2.0 leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General