# OSWorld 2.1 (offline) — AI model rankings

Can the AI actually operate a computer — clicking, typing, and using real apps — to finish tasks on its own? Version 2.1, in its offline setting. Scores are not comparable with version 2.0. Higher is better.

10 tracked models have a published OSWorld 2.1 (offline) score. Higher is better. Scores come from published lab reports and benchmark sources; recorded sources appear beside the scores.

## Ranking

| Rank | Model | Developer | Score | Source | Released |
| --- | --- | --- | --- | --- | --- |
| 1 | Claude Fable 5.2 | Anthropic | 89.3% | Lab | Nov 9 2026 |
| 2 | Claude Opus 5.5 | Anthropic | 87.2% | Lab | Sep 22 2026 |
| 3 | Claude Sonnet 5.5 | Anthropic | 83.9% | Lab | Sep 28 2026 |
| 4 | Claude Fable 5.1 | Anthropic | 83.5% | Lab | Sep 1 2026 |
| 5 | Gemini 4 Carbon | Google | 81.5% | Lab | Oct 28 2026 |
| 6 | GPT-6 Astra | OpenAI | 79.5% | Lab | Sep 3 2026 |
| 7 | Gemini 4 Hydrogen | Google | 77.6% | Lab | Nov 12 2026 |
| 8 | GPT-6.1 Terra | OpenAI | 74.2% | Lab | Oct 26 2026 |
| 9 | Claude Haiku 5.5 | Anthropic | 72.4% | Lab | Oct 7 2026 |
| 10 | Claude Haiku 4.5 | Anthropic | 15.7% | Lab | Oct 15 2025 |


---

Canonical page: https://aireleasetracker.com/benchmark/osworld-2.1-offline
Site index: https://aireleasetracker.com/llms.txt
Source: AI Release Tracker (https://aireleasetracker.com). Most benchmark scores come from lab launch material; gathered results identify the leaderboard that published them.
