Structured outputs for Anthropic Claude models on Amazon Bedrock

Amazon Bedrock now supports structured outputs for all Anthropic Claude models, enabling schema-compliant JSON responses through constrained decoding on both bedrock-runtime and bedrock-mantle endpoints.

• Constrained decoding enforces JSON schema during token generation, eliminating parsing errors and retry logic
• Two mechanisms available: JSON Schema output format for data extraction and strict tool use for agentic workflows
• Responses are type-safe with required fields always present and reduced output token count
• Compiled grammars are cached for 24 hours, improving performance for repeated requests
• Supports batch inference, cross-Region inference, and streaming of structured responses
• Works with all Anthropic Claude models including Haiku 5.5 for classification and extraction tasks
• Requires additionalProperties: false and supports JSON Schema Draft 2020-12 subset features

Structured outputs eliminate custom validation code, making production AI applications more reliable for data extraction, agent tool calls, and classification workflows.