OSWorld-Verified (Fable/Mythos): leaderboard
Computer-use agents complete tasks in web and desktop applications, manage files, and carry out workflows across apps. Execution-based checks measure task success (%). This slice reports Anthropic's own runs from the Claude Fable 5 / Mythos 5 system card; small frontier-model pool.
Metric: Score (%). Source: www-cdn.anthropic.com. Saturation forecast: Around January 2027. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Mythos Preview | 85.4 | #4 |
| 2 | Claude Fable 5 | 85 | #16 |
| 3 | Claude Mythos 5 | 85 | #2 |
| 4 | Claude Opus 4.8 | 83.4 | #37 |
| 5 | GPT-5.5 | 78.7 | #31 |
| 6 | Gemini 3.5 Flash | 78.4 | #60 |
| 7 | Gemini 3.1 Pro (Preview) | 76.2 | #58 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=osworld-verified-fable-mythos · How It Works · Data refreshed daily, snapshot 2026-10-20.