Atlas

Benchmarks

← All benchmarks

OSWorld 2.0 500-step Partial Score

Agents · 2026-06-25

OSWorld 2.0 official leaderboard partial score under a 500-step action budget.

Top models (higher is better)

ModelScore
Claude Opus 570.6
Claude Fable 566.1
GPT-5.6 Sol62.6
Opus 4.857.1
GPT-5.549.5
Opus 4.749.1
Sonnet 4.641.5
MiniMax M322.3
Kimi K2.622.1
Qwen3.7-Plus21.5
Loading Atlas data…