[Bug]: Qwen3 enable_thinking is broken when continue_final_message is true

[Huarong]

Your current environment

The output of python collect_env.py

Docker image vllm/vllm-openai:v0.11.0

🐛 Describe the bug

Server

docker compose command to deploy the server:

command: --model /data/model --served-model-name demo-gpt --host 0.0.0.0 --port 8000 -tp 2 --tokenizer-mode auto --trust-remote-code --gpu-memory-utilization 0.95 --max-model-len 8192 --max-num-seqs=100 --safetensors-load-strategy eager --enable-log-requests --enable-log-outputs --reasoning-parser qwen3

Request with continue_final_message is False

{
    "model": "qwen3-32b",
    "messages": [
        {
            "role": "system",
            "content": "You are an AI"
        },
        {
            "role": "user",
            "content": "1+1=?"
        }
    ],
    "chat_template_kwargs": {"enable_thinking": true},
}

prompt from server log:

prompt: '<|im_start|>system\nYou are an AI<|im_end|>\n<|im_start|>user\n1+1=?<|im_end|>\n<|im_start|>assistant\n'

Request with continue_final_message is True

{
    "model": "qwen3-32b",
    "messages": [
        {
            "role": "system",
            "content": "You are an AI"
        },
        {
            "role": "user",
            "content": "1+1=?"
        },
        {
            "role": "assistant",
            "content": "<think>\nGood Question"
        }
    ],
    "add_generation_prompt": false,
    "continue_final_message": true,
    "chat_template_kwargs": {"enable_thinking": true}
}

prompt from server log:

prompt: '<|im_start|>system\nYou are an AI<|im_end|>\n<|im_start|>user\n1+1=?<|im_end|>\n<|im_start|>assistant\n<think>\n\n</think>\n\n<think>\nGood Question

The Bug

When set {"enable_thinking": true} and "continue_final_message": true, extra <think>\n\n</think>\n\n was inserted to the prompt.

Before submitting a new issue...

• Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.