How is response format set to JSON Schema?

[alehuo]

I appreciate your work for the SDK. I'm currently testing the SDK and was wondering how one can configure the completions endpoint to emit JSON Schema. This is what I've currently tried:

from openrouter import OpenRouter
from openrouter.components import JSONSchemaConfig, ResponseFormatJSONSchema
with OpenRouter(api_key=self.config.api_key) as client:
            response = await client.completions.generate_async(
                model=self.config.model,
                prompt=prompt,
                seed=self.config.seed,
                temperature=self.config.temperature,
                top_p=self.config.top_p,
                response_format=ResponseFormatJSONSchema(
                    json_schema=JSONSchemaConfig(
                        name="Hello world",
                        description="Foo",
                        schema_=schema.model_json_schema(),
                    )
                ),
            )

[ArjunJagdale]

The Python SDK exposes ResponseFormatJSONSchema and JSONSchemaConfig because it mirrors the OpenAI SDK interface. However, the OpenRouter Completions API does not currently implement JSON Schema response formatting, so the parameter has no effect.

JSON Schema output works only with /chat/completions and only for models that natively support structured responses (e.g., OpenAI 4.1, 4o). For all other models, schema enforcement must be done via prompting.

[ArjunJagdale]

Let us say -

prompt = f"""
Your response MUST be valid JSON matching this schema:

{schema}