OSWorld (v1)

Original OSWorld computer-use tasks. Near saturation.

CategoryAgents
MetricSuccess rate
SourceEpoch AI (evaluator)
LicenseCC BY 4.0
Benchmark pageos-world.github.io
StatusScored
Weight0.5 within Agents
FitDiscrimination 0.95, noise 0.36

Results

#ModelProviderResultDate
1 Claude Sonnet 4.6 Anthropic 72.1% 2026-11-11
2 Claude Opus 4.5 Anthropic 66.3% 2026-11-11
3 Kimi K2.5 Moonshot AI 63.3% 2026-11-11
4 Claude Sonnet 4.5 Anthropic 62.9% 2026-11-11
5 Claude Sonnet 4.5 Anthropic 58.1% 2026-11-11
6 Claude Sonnet 4 Anthropic 43.9% 2026-11-11
7 Claude Sonnet 4.5 Anthropic 42.9% 2026-11-11
8 Claude Sonnet 4 Anthropic 41.4% 2026-11-11
9 Claude Sonnet 3.7 Anthropic 35.8% 2026-11-11
10 Claude Sonnet 3.7 Anthropic 35.6% 2026-11-11
11 computer-use-preview-2025-03-11 Unknown 31.3% 2026-11-11
12 Claude Sonnet 4 Anthropic 31.2% 2026-11-11
13 computer-use-preview-2025-03-11 Unknown 30.5% 2026-11-11
14 Claude Sonnet 3.7 Anthropic 27.1% 2026-11-11
15 computer-use-preview-2025-03-11 Unknown 26.0% 2026-11-11
16 o3 Medium OpenAI 23.0% 2026-11-11
17 o3 Medium OpenAI 17.2% 2026-11-11
18 o3 Medium OpenAI 9.1% 2026-11-11
19 Qwen2.5-VL 72B Instruct Alibaba 5.0% 2026-11-11
20 Qwen2.5-VL 72B Instruct Alibaba 4.4% 2026-11-11