Amazon Bedrock adds schema-enforced structured outputs for all Anthropic Claude models on both endpoints

Amazon Bedrock now supports structured outputs for all Anthropic Claude models, on both the bedrock-runtime and bedrock-mantle endpoints, wherever the model is available. According to AWS, responses can be constrained to a developer-defined JSON schema, which removes the need for custom parsing, retry logic and post-processing.

What shipped

AWS describes two mechanisms, available on both endpoints:

Feature	Purpose	Typical use
JSON Schema output format	Control response structure	Data extraction, report generation, API responses
Strict tool use	Validate tool parameters against schemas	Agentic workflows, function calling, multi-step automation

Structured outputs use constrained decoding, which enforces the schema during token generation rather than validating afterward. AWS says the model cannot emit tokens that would violate the schema. AWS describes this as moving output formatting from probabilistic to deterministic.

Per AWS, this means:

• Responses are valid JSON, so JSON.parse() errors should not occur.
• Field types are enforced and required fields are always present.
• Output is limited to the fields the schema defines. AWS says this reduces output token count and can lower inference costs on high-volume workloads. No figures are given.

How it works

Bedrock validates the schema against a supported subset of JSON Schema Draft 2020-12, then compiles a grammar artifact. The first request with a new schema may take longer because of this compilation step.

Compiled grammars are cached for 24 hours. The cache is scoped to the account and runs from first access. Changing the structure of the schema invalidates the cache and triggers recompilation. Changing only the name or description fields does not.

Integration details

• InvokeModel (bedrock-runtime): takes the Anthropic-native request body. The schema goes in output_config.format as a JSON object, not a JSON string, with type set to json_schema.
• Messages API (bedrock-mantle): strict tool use is enabled by setting "strict": true on a tool definition. The tool input then matches the input_schema exactly, including enum values and additionalProperties: false.
• Batch inference: supported, so large asynchronous jobs still return schema-compliant output.
• Cross-Region inference: supported on bedrock-runtime only, with no additional setup.
• Streaming: structured responses can be streamed as typed events.

AWS's sample code checks stop_reason before parsing. If it is max_tokens, the response was truncated and the developer must raise max_tokens and retry. Schema enforcement does not remove the need to size output budgets correctly.

The sample code uses model IDs such as global.anthropic.claude-sonnet-5-5 and anthropic.claude-sonnet-5. The post does not list supported models by version in the excerpt reviewed here.

AWS cites use cases including extracting fields from earnings reports and compliance documents, parsing clinical notes into EHR formats, enriching product catalogs, analyzing contracts, and routing support tickets.

Pricing for the feature itself is not stated in the post. Standard Claude token pricing on Bedrock presumably applies, but AWS does not confirm this in the material reviewed.

What this means

This is a platform parity update, not a new model. Anthropic and other providers have offered schema-constrained output through their own APIs, and Bedrock customers can now get the same behavior without leaving AWS. For teams running extraction or classification pipelines on Bedrock, retry and repair code for malformed JSON can likely be deleted.

The strict tool use support on Mantle matters more for agent builders. A malformed tool call can derail a multi-step workflow, and constrained decoding removes one class of that failure.

Two caveats apply. Schema compliance guarantees structure, not correctness: a field can be well-typed and still wrong. And the supported schema is a subset of Draft 2020-12, so complex schemas need testing against Bedrock's validator before production rollout.