Setting "enable_thinking": false in chat_template_kwargs for Qwen3.5/3.6 models disables the generation of reasoning/thinking content—the model will not produce or output any thinking tokens, and only the final answer is generated (docs.vllm.ai reasoning outputs, vllm-project/vllm#35574). If you still see reasoning content, ensure you are using the latest vLLM version and that the model/chat template supports this flag.
Would you like troubleshooting steps or a sample request?
Sources:
Hint: Mention @RunLLM in the post for followups.