OSWorld 2.0 leaderboard
- Claude Opus 5 leads
- 31.4%
- models measured
- 9
- lead over #2
- +4.1 pts
- median
- 13.0%
Leader on OSWorld 2.0 · read October 22
On October 22, 2026, 9 of our catalog's models had a OSWorld 2.0 result, and Anthropic Claude Opus 5 led with 31.4%, ahead of OpenAI GPT-5.6 Sol at 27.3%.
Tasks on a real desktop computer, done through the screen, mouse and keyboard. The number is the share of tasks completed.
Role in the index
OSWorld 2.0 is shown as published and does not feed any of our indexes. Read the methodology
| # | Model | Result | Estimated cost | With partial credit | Evidence | Coding index | Output $/Mtok |
|---|---|---|---|---|---|---|---|
| 1 | Claude Opus 5Anthropic | 31.4% | Not published | 68.3% | 83.0 | $25.00 | |
| 2 | GPT-5.6 SolOpenAI | 27.3% | Not published | 62.7% | 73.5 | $20.00 | |
| 3 | Claude Opus 4.8Anthropic | 20.6% | Not published | 54.8% | 65.0 | $25.00 | |
| 4 | Claude Opus 4.7Anthropic | 18.2% | Not published | 48.9% | 67.5 | $25.00 | |
| 5 | GPT-5.5OpenAI | 13.0% | $2750.00 | 49.5% | 71.3 | $30.00 | |
| 6 | Claude Sonnet 4.6Anthropic | 9.3% | $1550.00 | 33.9% | 57.3 | $15.00 | |
| 7 | MiniMax M3MiniMax | 4.6% | $258.78 | 22.3% | 39.2 | $1.20 | |
| 8 | Kimi K2.6Moonshot AI | 4.6% | $708.00 | 22.1% | 52.1 | $4.00 | |
| 9 | Qwen3.7 PlusQwen | 2.8% | $411.56 | 21.5% | 33.2 | $1.28 |
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. Estimated cost is the estimated spend on the run, in US dollars, as Epoch AI records it beside the result. Every other column is a figure the board publishes for the same run.