This is a mixture-of-experts model featuring hybrid thinking modes that allow you to toggle between quick responses and step-by-step reasoning. It’s optimized for coding, mathematics, and agentic workflows with support for 119 languages.
qwen-3-235b-a22b
model to make way for enhanced versions that deliver superior performance on reasoning tasks. We recommend migrating to either Qwen 3 235B Instruct or Qwen 3 235B Thinking.qwen-3-235b-a22b
/no_think
to your prompt to disable the model's default reasoning behavior.Tell me about cats /no_think
temperature=0.6
and top_p=0.95
, and avoid greedy decoding completely as it causes performance issues and repetitions.Tier | Requests/min | Input Tokens/min | Output Tokens/min | Daily Tokens |
---|---|---|---|---|
Free | 30 | 64k | 8k/request | 1M |
Chat Completions
Completions