Structured output via sglang not working with DeepSeek-V3.1

[yuyue0225sc]

Hi folks,

I ran into a strange issue when using sglang with different models.

When I start a DeepSeek-V3.1 service using sglang and make the following request:

import json
from openai import OpenAI

def main():
    niogpt_base_url = "xxxx"
    niogpt_api_key = "xxxx"
    client = OpenAI(
        api_key=niogpt_api_key,
        base_url=niogpt_base_url
    )

    json_schema = {
        "type": "object",
        "properties": {
            "name": {"type": "string", "pattern": "^[\\w]+$"},
            "population": {"type": "integer"},
        },
        "required": ["name", "population"],
    }

    response = client.chat.completions.create(
        model="DeepSeek-V3.1",
        messages=[
            {
                "role": "user",
                "content": "Give me the information of the capital of France in the JSON format.",
            }
        ],
        temperature=0,
        max_tokens=128,
        response_format={
            "type": "json_schema",
            "json_schema": {"name": "foo", "schema": json_schema},
        },
    )

    print(response.choices[0].message.content)


if __name__ == "__main__":
    main()

The output does not match the expected structured JSON format shown in the official example:

However, if I use the same code with a Qwen3-8B service:

import json
from openai import OpenAI

def main():
    client = OpenAI(
        api_key="any_key_you_like",
        base_url="http://0.0.0.0:30000/v1"
    )

    json_schema = json.dumps({
        "type": "object",
        "properties": {
            "name": {"type": "string", "pattern": "^[\\w]+$"},
            "population": {"type": "integer"},
        },
        "required": ["name", "population"],
    })

    response = client.chat.completions.create(
        model="qwen3-8B",
        messages=[{
            "role": "user",
            "content": "Give me the information of the capital of France in the JSON format.",
        }],
        temperature=0,
        max_tokens=128,
        response_format={
            "type": "json_schema",
            "json_schema": {"name": "foo", "schema": json.loads(json_schema)},
        },
    )

    print(response.choices[0].message.content)

if __name__ == "__main__":
    main()

The output matches the expected structured JSON format:

{
  "name": "Paris",
  "population": 2148244
}

So my question is:

Why does the official sglang structured output example work with Qwen3-8B model services but not with DeepSeek-V3.1 services?

Is there a difference in structured output support between these models, or is there a specific configuration needed for DeepSeek-V3.1?

Thanks in advance for any insights!

[jondurbin]

Seems to be the case with other models as well, 3.1 terminus as you might expect, but also GLM-4.5, GLM-4.6, etc.

[jondurbin]

chutesai#3 fixed it for us (chutes) on deepseek 3.1, 3.1-terminus