All tasks / Models
OpenCUA-72B
Open-weights computer-use agent models from XLANG Lab and Moonshot AI, in 7B, 32B and 72B sizes (MIT).
- XLANG Lab (The University of Hong Kong) & Moonshot AI
- Open weights
- Released Oct 2025
Where it ranks
- –Browser & computer agentsModels · Local · no result on this task’s benchmarks yet
Facts
- Vendor
- XLANG Lab (The University of Hong Kong) & Moonshot AI
- Family
- OpenCUA
- Released
- Oct 2025
- Weights
- Open: download and run it yourself
- License
- MIT
- Parameters
- 73.4B
- GPU memory to run locally
- ~43 GB (4-bit ≈ 73.4 B × 0.57 GB + 1.5 GB ≈ 43 GB, estimate (a 48 GB card); 32B and 7B variants fit consumer GPUs.)
- Access
- Free (open weights)
- Last verified
- 18 Oct 2026
Price from huggingface.co, 18 Oct 2026.
Benchmark results
| Benchmark | Result | Entry on the leaderboard | As of | Source |
|---|---|---|---|---|
| OSWorld-Verified (models) | 46.1% | opencua-72b-previewSpecialized model; 100 max steps; 165.89/360 tasks; entry dated 2025-10-01, the day the OpenCUA-72B weights were published (the board labels it 'preview'); read from osworld_verified_results.xlsx | 18 Oct 2026 | os-world.github.io |
Where you can use it
No tracked tool lists this model yet; as open weights it runs in local runtimes.