What changed
Amazon Bedrock announced structured outputs support for Anthropic Claude models. The model no longer produces tokens that would violate a supplied JSON schema. AWS calls this constrained decoding: the schema is enforced during generation rather than checked afterward.
Output formatting shifts from probabilistic to deterministic. Developers no longer need to handle JSON.parse() errors, missing fields or retry logic for schema violations.
Two mechanisms
- JSON Schema output format: Controls the response structure. Used for data extraction, report generation and API responses.
- Strict tool use: Validates tool parameters against schemas. Aimed at agentic workflows, function calling and multi-step automation.
How it works
Bedrock first validates the schema against a supported subset of JSON Schema Draft 2020-12. For new schemas it compiles a grammar artifact, so the first request may take slightly longer. Compiled grammars are cached for 24 hours and scoped to the account. Changing the schema structure invalidates the cache, while changing only name or description fields leaves it intact.
Use cases and limits
AWS cites extracting fields from financial reports, parsing clinical notes into electronic health records, contract analysis and support ticket routing. The feature works with batch inference, and cross-Region inference is supported on bedrock-runtime. For real-time needs, structured responses can be streamed as typed events. Because output is limited to schema-defined fields, token counts and inference costs can drop for high-volume workloads.
