QVQ-72B-Preview
Open 72B visual-reasoning preview on Qwen2-VL-72B scoring 70.3 on MMMU; long reasoning applied to images.
Extends the reasoning-model idea (QwQ) to visual inputs. Released under the Qwen license per Wikipedia; QVQ-Max-Preview followed on 2025-03-28.
- Date
- Wednesday, 25 December 2024
- Lab
- Alibaba (Qwen)
- Kind
- open-weights
- Access
- open weights (restricted license)
Figures
| Measure | Value | Measured by |
|---|---|---|
| MMMU | 70.3 Alibaba-reported; large gains on math benchmarks vs Qwen2-VL-72B-Instruct | company |
Blog dated 2024-12-25. The model is experimental, and the post warns of language mixing and circular reasoning. License per Wikipedia.
Sources
This record was checked against its sources on 6 October 2026. How we check