OSWorld leaderboard
- Claude Sonnet 4.6 leads
- 72.1%
- models measured
- 6
- lead over #2
- +5.8 pts
- median
- 63.1%
Leader on OSWorld · read October 22
On October 22, 2026, 7 of our catalog's models had a OSWorld result, and Anthropic Claude Sonnet 4.6 led with 72.1%, ahead of Anthropic Claude Opus 4.5 at 66.3%.
The first OSWorld set: tasks on a real desktop computer, done through the screen, mouse and keyboard. Share of tasks completed.
Role in the index
OSWorld is shown as published and does not feed any of our indexes. Read the methodology
| # | Model | Result | Evidence | Coding index | Output $/Mtok |
|---|---|---|---|---|---|
| 1 | Claude Sonnet 4.6Anthropic | 72.1% | 57.3 | $15.00 | |
| 2 | Claude Opus 4.5Anthropic | 66.3% | 44.0 | $25.00 | |
| 3 | Kimi K2.5Moonshot AI | 63.3% | 38.6 | $2.85 | |
| 4 | Claude Sonnet 4.5Anthropic | 62.9% | 39.5 | $15.00 | |
| 5 | Claude 3.7 SonnetRetiredAnthropic | 35.8% | Not benchmarked | $15.00 | |
| 6 | o3OpenAI | 23.0% | Not benchmarked | $8.00 | |
| 7 | Qwen2.5 VL 72B InstructQwen | 5.0% | Not benchmarked | $1.00 |
Only models in our catalog are listed, and each appears once with its best published run. 215 rows were read from the source on October 22, 2026, and rows that match no catalog model are left out.