Amazon Bedrock adds schema-enforced structured outputs for every Anthropic Claude model

AWS has extended structured outputs to every Anthropic Claude model on Amazon Bedrock, across both the bedrock-runtime and bedrock-mantle endpoints. Developers can supply a JSON Schema as the output format, or enable strict tool use, and receive responses guaranteed to conform. Teams no longer need the repair loops that parse model output, detect malformed JSON and re-prompt.

Mechanically, constrained output means the model's generation is held to the schema rather than validated after the fact. For agent pipelines this removes a whole class of failures: a missing field or a stray prose preamble before the JSON that crashes a downstream step. Strict tool use applies the same guarantee to function-call arguments, which matters most when agents chain many tools and one malformed call can derail a long run.

The feature brings Bedrock's Claude offering in line with what developers expect from first-party APIs, where structured outputs have become standard. It also fits AWS's broader push this month to make Bedrock the default enterprise agent platform. That push includes TwelveLabs' Pegasus 1.5 video model, which emits timestamped JSON, Grok 4.7 in GovCloud, and earlier October previews of Bedrock Managed Agents and the Well-Architected Agent lens.

Developer reaction has been positive, with many noting the end of hand-rolled JSON parsing and retry logic. Caveats remain: schema enforcement guarantees shape, not truth, so a schema-valid object can still contain hallucinated values. Very complex or deeply recursive schemas may also carry latency or support limits. Teams should still validate business logic, but they can drop the brittle repair code.