Qwen-VLA (vision-language-action generalist)
Unified vision-language-action model on Qwen3.5-4B plus a 1.15B flow-matching action decoder, covering manipulation, navigation and trajectory prediction.
One generalist trained jointly on robot manipulation, human egocentric, simulation and navigation data. Qwen says it matches or beats specialists fine-tuned per benchmark. The June Qwen-Robot Suite ships separate navigation and manipulation models.
- Date
- Thursday 28 May 2026
- Lab
- Alibaba (Qwen)
- Kind
- model
- Access
- open weights
Figures
| Measure | Value | Measured by |
|---|---|---|
| Technical report | arXiv 2605.30280 Revised v2 on 2026-06-01 | company |
Date is GitHub repo creation (2026-05-28). The repository is public, but it is not clear that weights are posted or under what license. Benchmark claims are Qwen's own and have not been independently verified.
Sources
This record was checked against its sources on 6 October 2026. How we check