QwQ Models by Alibaba
Last refreshed 2026-09-09. Next refresh: weekly.
Details
Capabilities
About
QwQ is the reasoning model line in Alibaba's Qwen series, made by the Qwen Team. Unlike standard instruction-tuned models, QwQ thinks through a problem before it answers, which is aimed at hard math, coding, and other multi-step tasks. QwQ-32B, the medium-sized model, is built on Qwen2.5-32B, has 32.5B parameters, and supports a 131,072-token context window. It was trained with reinforcement learning on math and coding, checked by an answer verifier and a code-execution server, followed by a shorter RL stage for general skills. It can also use tools and adjust its reasoning based on feedback. Weights are open on Hugging Face and ModelScope.
Decision facts
- Best fit
- reasoningcodingmath-heavy prompts
- Capability starting point
- QwQ 32B with 128K context and reasoning
- Lowest tracked input
- QwQ 32B · $0.660/1M · Cloudflare Workers AI
- Closest related family
- Ovis Image
Current Variants
Use-when guidance is based on each model's tracked capabilities, context window, release date, and replacement status.
Use when the workload needs reasoning, 128K context, and 32.5B parameters.
| Model | Use when | Released | Signals | Status |
|---|---|---|---|---|
| QwQ 32B | Use when the workload needs reasoning, 128K context, and 32.5B parameters. | 2025-03 | reasoning128K context32.5B parameters | Current |
Release Timeline
1 release groupSpecifications(1 models)
| Model | Released | Context | Parameters | Reasoning |
|---|---|---|---|---|
| QwQ 32B | 2025-03 | 128K | 32.5B | Yes |
Available From(2 providers)
Pricing
| Model | Provider | Input / 1M | Output / 1M | Type |
|---|---|---|---|---|
| QwQ 32B | Cloudflare Workers AI | $0.66 | $1 | Serverless |
