OSWorld 2.0 500-step Binary Accuracy
Agents · 2026-06-25
OSWorld 2.0 official leaderboard binary completion accuracy under a 500-step action budget.
Top models (higher is better)
| Model | Score |
|---|---|
| Opus 4.8 | 20.6 |
| Opus 4.7 | 18.2 |
| GPT-5.5 | 13.0 |
| Sonnet 4.6 | 9.3 |
| Kimi K2.6 | 4.6 |
| MiniMax M3 | 4.6 |
| Qwen3.7-Plus | 2.8 |