Atlas

Benchmarks

← All benchmarks

OSWorld 2.1 full - reported binary completion, step limit unspecified

Agents

Strict binary completion on the full 108-task OSWorld 2.1 set, with unreported step budget. It shares a factor group with the fully specified full-set protocol rather than supplying a second independent task corpus.

Top models (higher is better)

ModelScore
Muse Wonder Full39.4
Muse Wonder34.3
Loading Atlas data…