QwQ 32B
by Alibaba · qwen-3 family
32B
parameters
text-generation code-generation reasoning multilingual math
QwQ 32B is Alibaba's reasoning-focused model from the Qwen family, designed to excel at complex mathematical and logical reasoning tasks through chain-of-thought processing. At 32B parameters, it delivers reasoning performance that punches well above its weight class. The model is particularly strong at math competitions, code reasoning, and scientific problem solving. Its relatively compact size makes it accessible on high-end consumer GPUs at Q4 quantization, offering an excellent balance of reasoning capability and hardware requirements.
Quick Start with Ollama
ollama run q4_K_M | Creator | Alibaba |
| Parameters | 32B |
| Architecture | transformer-decoder |
| Context | 128K tokens |
| Released | Mar 6, 2025 |
| License | Apache 2.0 |
| Ollama | qwq |
Quantization Options
| Format | File Size | VRAM Required | Quality | Ollama Tag |
|---|---|---|---|---|
| Q4_K_M rec | 18.5 GB | 21.5 GB | | q4_K_M |
| Q8_0 | 34 GB | 37 GB | | q8_0 |
| F16 | 64 GB | 68 GB | | fp16 |
Compatible Hardware
Q4_K_M requires 21.5 GB VRAM
Benchmark Scores
82.5
mmlu