QwQ 32B

Permissive

Alibaba · Qwen · Released 2025-03

32BText

A reasoning-tuned model that produces long chain-of-thought traces before answering, aimed at math and logic tasks rather than general chat.

Strengths

  • +Apache 2.0 license
  • +Strong math and logic benchmark scores for its size
  • +Competitive with much larger reasoning models

Limitations

  • -Verbose chain-of-thought output increases latency and token cost
  • -Can overthink simple prompts relative to a plain instruct model

License

Apache 2.0

use commercially with attribution niceties

Hardware

wants 24-48GB of VRAM - a 3090/4090-class card or better

Links

Stats

Qwen/QwQ-32B— downloads·— likes

via Hugging Face

Related models