Qwen2.5 32B Instruct
Qwen · released 2024-09-17 · apache-2.0 license
Qwen2.5-32B-Instruct is the strong Apache-2.0 mid-size Qwen2.5 model: vendor GPQA 49.5, MMLU-Pro 69.0, LiveCodeBench 51.2, reported to surpass the larger Qwen2-72B.
Key specs
| Type | Local open-weight |
|---|---|
| Parameters | 32.76B total |
| Architecture | qwen2 |
| Context window | 33K tokens |
| Knowledge cutoff | — |
| Modalities | text |
| Recommended backends | — |
| Minimum viable rig | — |
Benchmark scores
| GPQA Diamond | 49.5% |
|---|---|
| SWE-bench Verified | — |
| AIME | — |
| MMLU-Pro | 69% |
| BFCL v3 (tool use) | — |
| Composite score | 6.22 |
| Community rating | No reviews yet |
VRAM & disk per quantization
| Quant | VRAM | Disk | RAM | Context |
|---|---|---|---|---|
| Q4_K_M | 20.5 GB | 19 GB | — | 33K |
Strengths & weaknesses
Strengths: Surpasses the larger Qwen2-72B on most vendor evals; Strong math (vendor MATH 83.1); Apache-2.0
Weaknesses: No native reasoning mode; No vendor SWE-bench/AIME numbers