ai-agents-structured-output-co-using-a-pydantic-model-as-response-format-or-with--9f25ba16

THEN Replace response_format with {'type': 'json_object'} and add instructions in the system message to output JSON matching the expected schema. Alternatively, use a model like Llama 3.3 that supports json_schema.

from litellm import completion
response = completion(
    model='deepseek-chat',
    messages=[{'role': 'system', 'content': 'Output JSON with fields: topic (string), subtopics (list of strings)'}, {'role': 'user', 'content': '...'}],
    response_format={'type': 'json_object'}
)