Computer Use Agent Navigation on OSWorld (Verified)
78.7Success RateGPT-5.5
Evaluation Results
| Method | Links | |
|---|---|---|
| GPT-5.5Model Type=Proprietary2026.05 | 78.7 | |
| Claude Opus 4.7Model Type=Proprietary2026.05 | 78 | |
| Kimi-K2.6Model Type=Open-source2026.05 | 73.1 | |
| Claude Sonnet 4.6Model Type=Proprietary2026.05 | 72.9 | |
| CUA-GYM-A17BTraining=RLVR on CUA-GYM, Base Model=Qwen3.5-397B-A17B2026.05 | 72.6 | |
| Qwen3.5-397B-A17BModel Type=Base Model2026.05 | 62.2 | |
| CUA-GYM-A3BTraining=RLVR on CUA-GYM, Base Model=Qwen3.5-35B-A3B2026.05 | 62.1 | |
| EvoCUA-32BModel Type=Open-source2026.05 | 56.7 | |
| Qwen3.5-35B-A3BModel Type=Base Model2026.05 | 54.5 | |
| EvoCUA-8BModel Type=Open-source2026.05 | 46.1 | |
| OpenCUA-72BModel Type=Open-source2026.05 | 45 | |
| Step-GUI-8BModel Type=Open-source2026.05 | 40.2 | |
| OpenCUA-32BModel Type=Open-source2026.05 | 34.8 |