OSWorld 2.1 Full, 500-step Partial Score
Agents · 2026-09-16
OSWorld 2.1 full 108-task release, partial completion credit with 500 steps.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Sonnet 5.5 | 80.1 |
| Claude Opus 5 | 77.7 |
Benchmarks
252 sources · 1,069 models × 2,256 benchmarks · 113,799 scoresAgents · 2026-09-16
OSWorld 2.1 full 108-task release, partial completion credit with 500 steps.
| Model | Score |
|---|---|
| Claude Sonnet 5.5 | 80.1 |
| Claude Opus 5 | 77.7 |