Benchmarks
OSWorld
Models scored
20
Best score
72.1
Metric
Score
OSWorld is an AI evaluation tracked by Epoch AI, with 20 scored model versions on record. The scores here are reported by third parties rather than produced by Epoch AI running the evaluation itself.
Scored models span September 2024 to February 2026. The best score recorded is 72.1, measured as Score. Models come from Anthropic, OpenAI, Alibaba, Moonshot.
Full record
- Score metric
- Score
- Models scored
- 20
- Best score recorded
- 72.1
- Earliest model scored
- Sep 19, 2024
- Most recent model scored
- Feb 17, 2026
- Third-party reported
- Yes
Leaderboard
| 01 | claude-sonnet-4-6 | Anthropic | 72.1 |
| 02 | claude-opus-4-5-20251101 | Anthropic | 66.3 |
| 03 | kimi-k2.5 | Moonshot | 63.3 |
| 04 | claude-sonnet-4-5-20250929 | Anthropic | 62.9 |
| 05 | claude-sonnet-4-20250514 | Anthropic | 43.9 |
| 06 | claude-3-7-sonnet-20250219 | Anthropic | 35.8 |
| 07 | computer-use-preview-2025-03-11 | 31.3 | |
| 08 | o3-2025-04-16_medium | OpenAI | 23 |
| 09 | Qwen2.5-VL-72B-Instruct | Alibaba | 5 |
More benchmarks
SourceEpoch AI, 'AI Benchmarking Hub'. Published online at epoch.ai. Retrieved 2026-07-29 from https://epoch.ai/benchmarks. Licensed under CC BY 4.0.
← All benchmarks