QwQ-32B-Preview
32B open-weights reasoning preview with 32K context, scoring 65.2% GPQA, 50.0% AIME and 90.6% MATH-500. It arrived a week after R1-Lite-Preview.
An open-weights o1-style model that reflects on its own reasoning ('Qwen with Questions'). Alibaba lists limits openly, including language mixing, recursive reasoning loops and safety caveats. It preceded March 2025's QwQ-32B, which added large-scale RL.
- Date
- Thursday, 28 November 2024
- Lab
- Alibaba (Qwen)
- Kind
- open-weights
- Access
- open weights
Figures
| Measure | Value | Measured by |
|---|---|---|
| GPQA / AIME / MATH-500 | 65.2% / 50.0% / 90.6% Alibaba-reported scores for QwQ-32B-Preview | company |
Blog dated 2024-11-28 (Beijing); Wikipedia cites a TechCrunch article from 2024-11-27. 'Better than o1-preview on some benchmarks' is the press characterisation of Alibaba's comparison chart, which I did not open.
Sources
This record was checked against its sources on 6 October 2026. How we check