benchgap
Agentic · tools

OSWorld 2.0 leaderboard

As of 2026-11-21, the highest measured score on OSWorld 2.0 is 79.6% by Kimi K3.1 Pro. 69 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#ModelScoreSource
1Kimi K3.1 Pro79.6%measured
2DeepSeek V4.1 Flash 101776.9%estimated ± 14.7 pp, low confidence
3Grok 4.776.9%estimated ± 14.7 pp, low confidence
4GPT-6 Astra72.6%measured
5GPT-6.2 Astra72.5%estimated ± 14.7 pp, low confidence
6Claude Opus 570.6%measured
7Gemini 4 Argon69.2%measured
8GPT-6.2 Sol67.1%estimated ± 14.7 pp, low confidence
9Muse Spark 1.366.9%measured
10GPT-5.6 Sol62.6%measured
11MiniMax M3.1 Flash61.7%measured
12GPT-6 Sol60.5%measured
13Mistral Large 460.3%estimated ± 14.7 pp, low confidence
14Gemini 3.8 Flash59.0%measured
15GLM-5.3-Flash59.0%estimated ± 5.5 pp, low confidence
16GPT-6 Luna56.6%estimated ± 14.7 pp, low confidence
17GPT-5.6 Terra50.2%measured
18Claude Opus 5.548.7%measured
19Gemini 3.7 Flash47.9%measured
20GPT-5.6 Luna45.6%measured
21Claude Sonnet 5.545.2%estimated ± 5.5 pp, low confidence
22Claude Fable 5.141.7%measured
23Beam 1.0 501B-A23B37.1%estimated ± 14.7 pp, low confidence
24Claude Opus 4.820.6%measured
25Claude Fable 519.9%estimated ± 3.6 pp, high confidence
26Claude Mythos 519.9%estimated ± 3.6 pp, high confidence
27ERNIE 5.5 Flash Preview19.6%estimated ± 5.5 pp, low confidence
28DeepSeek V4 Pro19.4%estimated ± 5.5 pp, low confidence
29Hy4 preview19.4%estimated ± 5.5 pp, low confidence
30Step 5 Preview19.4%estimated ± 5.5 pp, low confidence
31Qwen3.8-Flash-Next19.4%measured
32Qwen3.8 Max19.4%measured
33Kimi K319.4%estimated ± 5.5 pp, low confidence
34GLM-5.319.4%estimated ± 5.5 pp, low confidence
35Muse Spark Open19.4%estimated ± 5.5 pp, low confidence
36Ornith-1.5-397B19.4%estimated ± 5.5 pp, low confidence
37DeepSeek V4 Flash 073119.4%estimated ± 5.5 pp, low confidence
38Qwen3.8-27B19.2%estimated ± 3.6 pp, high confidence
39Claude Opus 4.718.2%measured
40Gemini 3.6 Flash17.8%estimated ± 3.6 pp, high confidence
41Holo3-35B-A3B17.3%estimated ± 3.6 pp, high confidence
42MiMo-V2.6-Pro16.7%estimated ± 3.6 pp, high confidence
43GLM-5.216.1%estimated ± 6.9 pp, low confidence
44Claude Sonnet 515.9%estimated ± 3.6 pp, high confidence
45MiMo-V2.6-Flash15.5%estimated ± 3.6 pp, high confidence
46Muse Spark 1.114.2%measured
47Holo3-122B-A10B13.4%estimated ± 3.6 pp, high confidence
48GPT-5.513.0%measured
49Gemini 3.5 Flash12.9%estimated ± 3.6 pp, high confidence
50Qwen3.7 Max11.5%estimated ± 5.9 pp, low confidence
51UI-Mate-27B11.4%estimated ± 3.6 pp, high confidence
52Gemini 3 Pro10.7%estimated ± 5.9 pp, low confidence
53MiMo-V2.5-Pro10.3%estimated ± 5.9 pp, low confidence
54GPT-5.49.3%estimated ± 3.6 pp, high confidence
55GLM-5.18.3%estimated ± 5.9 pp, low confidence
56Claude Sonnet 4.68.3%measured
57Gemini 3.5 Flash-Lite8.3%estimated ± 3.6 pp, high confidence
58Claude Opus 4.66.9%estimated ± 3.6 pp, high confidence
59GPT-5.4 mini6.2%estimated ± 3.6 pp, high confidence
60Gemini 3.1 Pro5.9%estimated ± 5.9 pp, low confidence
61Gemini 3 Flash5.7%estimated ± 5.9 pp, low confidence
62Kimi K2.64.6%measured
63MiniMax M34.6%measured
64Qwen3.6-27B4.4%estimated ± 5.9 pp, low confidence
65MiMo-V2-Pro3.2%estimated ± 6.9 pp, low confidence
66Qwen3.7 Plus2.8%measured
67Grok 4.12.3%estimated ± 6.9 pp, low confidence
68GPT-5.2-Codex2.1%estimated ± 5.9 pp, low confidence
69Step 3.7 Flash1.9%estimated ± 5.9 pp, low confidence
70Nemotron 3 Ultra1.8%estimated ± 14.7 pp, low confidence
71GLM-51.5%estimated ± 5.9 pp, low confidence
72Qwen3.6 Plus1.2%estimated ± 5.9 pp, low confidence
73GPT-5.1-Codex0.5%estimated ± 5.9 pp, low confidence
74Grok Build 0.10.1%estimated ± 5.9 pp, low confidence
75Claude Opus 4.50.1%estimated ± 3.6 pp, medium confidence
76Claude Sonnet 4.50.0%estimated ± 3.6 pp, medium confidence
77Gemini 2.5 Pro0.0%estimated ± 5.9 pp, low confidence
78GPT-5.10.0%estimated ± 5.9 pp, low confidence
79GPT-5.20.0%estimated ± 5.9 pp, low confidence
80GPT-5.3 Codex0.0%estimated ± 3.6 pp, medium confidence
81Grok 40.0%estimated ± 5.9 pp, low confidence
82Grok 4.1 Fast0.0%estimated ± 5.9 pp, low confidence
83Grok 4.30.0%estimated ± 5.9 pp, low confidence
84Kimi K2.50.0%estimated ± 3.6 pp, medium confidence
85MiMo-V2.50.0%estimated ± 5.9 pp, low confidence
86Muse Glimmer 30B0.0%estimated ± 3.6 pp, medium confidence
87Qwen3.5-122B-A10B0.0%estimated ± 3.6 pp, medium confidence
88Qwen3.5-27B0.0%estimated ± 3.6 pp, medium confidence
89Qwen3.5-35B-A3B0.0%estimated ± 3.6 pp, medium confidence
90Qwen3.5 397B0.0%estimated ± 3.6 pp, medium confidence
91Qwen3.6-35B-A3B0.0%estimated ± 5.9 pp, low confidence
92Qwen3 Max0.0%estimated ± 5.9 pp, low confidence
93UI-Mate-9B0.0%estimated ± 3.6 pp, medium confidence

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General