What are structured outputs?

Structured outputs constrain a model to emit valid JSON matching a schema you provide, instead of free text that your code must parse and hope over. The decoder only samples tokens that keep the output schema-valid, so downstream systems receive data they can trust mechanically.

This closes the gap between language models and software: extraction pipelines, form fillers, and agents all need fields, not prose. Schema-constrained decoding removes the retry-and-parse loop that unreliable JSON used to require.

The practical guidance is to keep schemas shallow and required fields few; deeply nested optional structures push models toward degenerate completions. Validate with real traffic before trusting any schema at volume.

Related terms