OpenCUA-72B
XLANG Lab · 2025-10-03 · 73.4B parameters
OpenCUA-72B is an MIT-licensed open-weight end-to-end computer-use foundation model released as the largest OpenCUA checkpoint and fine-tuned from Qwen2.5-VL-72B-Instruct on AgentNet and other GUI-agent and grounding datasets. The exact checkpoint has 73,410,777,344 parameters, uses three screenshots of visual history with reflective long chain-of-thought in the documented agent setup, and at release achieved a 45.0% average success rate on OSWorld-Verified, which its developers reported as the best among open-source models.
Benchmark scores
| Benchmark | Score |
|---|---|
| OSWorld-Verified | 45.0 |