Atlas

Benchmarks

← All benchmarks

OSWorld-Verified (Anthropic 361-task cohort)

Agents

The 361-task OSWorld-Verified cohort reported for Mythos 5, Mythos Preview and Opus 4.8 in Figure 8.16.3.A of Anthropic's Fable 5 and Mythos 5 system card. The figure reports pass@1 averaged over five runs, with 100 steps and max reasoning effort. Exact excluded task identities are not established, so this cohort is kept distinct from both the 369-task benchmark and other 361-task subsets.

Top models (higher is better)

ModelScore
Claude Mythos Preview85.4
Claude Fable 585.0
Opus 4.883.4
Loading Atlas data…