With a community of over one million developers who build FAST, Groq can’t help but want to keep up. That’s why we ship fast, like today’s launch of Qwen-qwq-32b on GroqCloud™.
Performance
The 32B parameter model is currently running at ~400 T/s and supports both tool calling and JSON mode on GroqCloud. Stay tuned for official 3rd party benchmarks from Artificial Analysis.
Pricing
Qwen-qwq-32b is available on Groq for $0.29/M input tokens and $0.39/M output tokens. You can explore our pricing information here.
Comparisons
QwQ 32B is the latest reasoning model of the Qwen series. By leveraging reinforcement learning, it delivers increased reasoning and intelligence compared to conventional pretraining and post-training methods. At just 32B parameters, QwQ 32B provides on-par performance with significantly larger reasoning models like the 671B parameter DeepSeek-R1.
Based on benchmarks published by the Qwen team, QwQ 32B is extremely proficient in mathematical reasoning, coding, and general problem solving. It delivers on-par performance with DeepSeek-R1 on both AIME24 (79.5%) and IFEval (83.9%) as well as outperforms DeepSeek-R1 on LiveBench (73.1%) and BFCL (66.4%). View additional results here.
Build Fast Now
Self-serve access to GroqCloud™ Developer Tier is live! Plus, an improved console, docs, and more. Take your build to the next level with higher rate limits as well as access to Batch and Flex Processing.You can learn about more benefits in our blog.
