Structured output
A model feature that constrains output to a specific JSON schema, making LLM responses safely parseable by downstream code.
Without structured output, agents have to ask the LLM nicely to return JSON and hope it complies. It usually does, but the occasional drift breaks the pipeline.
Structured output enforces the schema at decoding time — the model can only emit tokens that produce valid JSON conforming to the schema. The reliability difference at scale is the difference between "ships" and "doesn't".
All major model APIs in 2026 support structured output. The remaining art is writing schemas that fit how the model wants to think rather than how a backend engineer would design them.
Frequently asked
Is structured output the same as function calling?+
Closely related but not identical. Function calling is structured output applied to a "what tool should I call" schema. Structured output is the general primitive.