​
Streaming

For compatible chat requests, set stream: true to receive output as it is generated. Consume the stream and close the connection when finished. Streaming can reduce the wait for the first visible text.

​
Image understanding

Choose a model with image input support, then supply image_url content in the messages format accepted by that endpoint. Image pricing depends on the model; check Models and pricing.

​
Tool calls

Define tools in the request. When the model returns tool_calls, your application executes the allowed tool and sends its result back to the model. The model’s requested call does not itself execute your code.

​
JSON output

For models supporting JSON mode, set response_format to { "type": "json_object" } and explicitly request JSON in the system prompt. Validate the result before using it in your application.

​
Length, temperature and reasoning

Set an output limit appropriate to the task. Temperature controls randomness where supported. Reasoning parameters and context limits are model-specific; do not assume every model accepts the same options. See API overview, Authentication and Gateway configuration.