JSON mode

Request JSON output with response_format in Chat Completions or text.format in the Responses API. Use json_object for an object, or json_schema to describe the output’s structure.

Request an object

{
  "model": "deepseek-v4.1-flash",
  "messages": [
    {
      "role": "user",
      "content": "Return a JSON object with city Jakarta and country Indonesia."
    }
  ],
  "response_format": {
    "type": "json_object"
  },
  "stream": false
}

Request a schema

Chat Completions requires a nonempty schema name, an object schema, and an optional boolean strict.

{
  "response_format": {
    "type": "json_schema",
    "json_schema": {
      "name": "location",
      "strict": true,
      "schema": {
        "type": "object",
        "properties": {
          "city": {
            "type": "string"
          },
          "country": {
            "type": "string"
          }
        },
        "required": [
          "city",
          "country"
        ],
        "additionalProperties": false
      }
    }
  }
}

Responses API

For POST /v1/responses, put the format inside text.format. Schema fields sit directly inside the format, without a json_schema wrapper.

{
  "model": "deepseek-v4.1-flash",
  "input": "Return a JSON object with city Jakarta and country Indonesia.",
  "text": {
    "format": {
      "type": "json_object"
    }
  },
  "stream": false
}

Model support and validation

JSON support depends on the selected model. Some models use prompt instructions rather than constrained decoding; for those models, schema compliance is best effort even with strict: true. Check model capabilities and validate the returned JSON against your schema.

Enable the response-healing plugin for automatic JSON repair on non-streaming JSON requests. Always parse and validate the result in your application: syntactically valid JSON can still contain incorrect values or the wrong structure. Streaming JSON is not repaired.