OSWorld 2.1 full - reported partial reward, step limit unspecified
Agents
Mean partial reward on the full 108-task OSWorld 2.1 desktop workflow set. The release does not specify a step limit; never treated as the explicitly 500-step protocol.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Opus 5.5 | 81.8 |
| Muse Wonder Full | 78.4 |
| Muse Wonder | 72.6 |