InferenceInference performance, quantization and cost optimization
结构化输出
Making the model output strictly in predefined formats such as JSON.
Structured output guarantees parseable results via constrained decoding or validation—essential for agents and programmatic use: function arguments and tool results depend on stable JSON.