Response Format
Domain: Models & Providers
Motivation
Response format is how the model’s output is shaped — free-form text, reasoning blocks, and tool calls. On the DeepSeek Messages route, reasoning, text, and raw-string tool arguments are translated into harness chunks; GenerateOptions sampling is limited to temperature, maxTokens, and stop, with no tool_choice or structured-output field in the core vocabulary.
Concrete Example
Messages returns reasoning, text, and raw-JSON-string tool-call arguments as stream chunks; tool arguments must be valid JSON objects when newly generated.
Analogy
Choosing whether the answer is a paragraph, a numbered list, or a form to fill in.