Structured Output

An LLM capability that constrains model output to conform to a predefined schema, typically JSON or XML, enabling reliable programmatic consumption of responses. Supported natively by OpenAI, Anthropic, and others via API parameters that enforce valid JSON matching a JSON Schema. Research shows strict format constraints can degrade reasoning quality, so a best practice is to include a 'reasoning' field before answer fields in the schema.