Claude on Amazon Bedrock can now return data in a strict, fixed format

Amazon Bedrock, the AWS service for using AI models, now lets Anthropic's Claude models return answers in a strict, fixed format. Every response follows a layout the developer defines in advance. That means fewer garbled results when an AI tool fills in a CRM, an intake form or a support ticket.

What changed

Software needs tidy, predictable data. Until now, an AI model could sometimes return a messy answer that the app could not read, forcing extra checks or a second try. Structured outputs remove that problem. The feature covers all Claude models on both Bedrock endpoints (the two ways developers connect), wherever the model is offered. It uses two tools: a JSON Schema output format and strict tool use.

How it works

The format is enforced while the AI writes, not checked afterward. Bedrock uses a method called constrained decoding, so the model cannot produce text that breaks the layout. AWS says responses are valid JSON, field types are enforced and required fields are always present. Developers should not need custom parsing, retry logic or cleanup code.

Strict tool use applies to the bedrock-mantle endpoint, which is built for AI agents that take actions. It guarantees that the details an agent passes to a tool match the expected layout exactly.

What it is good for

AWS gives a simple example: reading a customer email and pulling out the name, email address, plan of interest and whether a demo was requested. Other uses it lists include:

• Extracting fields from earnings reports and compliance documents
• Turning clinical notes into electronic health records
• Pulling key terms, parties and obligations from contracts
• Routing support tickets by intent, sentiment and entities

It also works with batch jobs, streaming and, on the bedrock-runtime endpoint, cross-Region inference. Because answers hold only the fields the layout defines, AWS says output size shrinks, which can lower costs for high-volume work. The material gives no pricing details.

What to watch

The first request with a new layout can be slower, because Bedrock has to prepare it. Later requests with the same layout run with minimal overhead for 24 hours, and the cache is tied to your account. Changing the structure resets this. Changing only a name or description does not.

One more catch: a response can still be cut short if the length limit is set too low. AWS's own example checks for that before using the result.