Qwen2.5-Coder series (up to 32B)
Full Qwen2.5-Coder family from 0.5B to 32B; the 32B-Instruct is billed as SOTA open code model, matching GPT-4o and scoring 73.7 on Aider.
Completes the Coder line announced in September with the 32B flagship and all intermediate sizes. Wikipedia lists the 32B, 14B, 7B, 1.5B and 0.5B under Apache 2.0 and 3B under the Qwen Research license.
- Date
- Tuesday, 12 November 2024
- Lab
- Alibaba (Qwen)
- Kind
- open-weights
- Access
- open weights
Figures
| Measure | Value | Measured by |
|---|---|---|
| Aider code repair (32B-Instruct) | 73.7 Alibaba says comparable to GPT-4o; also best open model on EvalPlus, LiveCodeBench and BigCodeBench (company) | company |
Blog dated 2024-11-12; Wikipedia dates the series to December 2024. Licenses are per Wikipedia. Scores are Alibaba-run.
Sources
This record was checked against its sources on 6 October 2026. How we check