Bedrock can now constrain generation to a JSON Schema, so a response that violates the schema cannot be produced. What it supports, what it does not (recursion, numeric and string constraints), the first-call latency, and how it changes extraction pipelines and agents.
Getting JSON out of a model used to mean asking nicely, parsing, validating and retrying when it failed. Structured outputs on Bedrock, launched in February 2026, use constrained decoding: the model cannot emit a token that would break your schema. In October 2026 AWS extended support to every Claude model on Bedrock's runtime endpoints, alongside models from OpenAI, Mistral, Qwen, DeepSeek and others.
JSON Schema output format: the whole response conforms to a schema you pass. For extraction, classification and any response your code parses.
Strict tool use: tool arguments conform to the tool's declared input schema. For agents, where a malformed argument used to mean a failed call or a retry.
Both work through the Converse and InvokeModel APIs.
Book a free 30-minute call. You'll leave knowing what we'd do first and roughly what it costs, whether or not you hire us.
The gaps matter. A schema can guarantee that "amount" is a number, not that it is positive. Keep a validation step for business rules, and treat structured outputs as removing format failures, not logic failures.
The first request with a new schema compiles it into a grammar, which adds latency. Compiled grammars are cached for 24 hours per account. Keep schemas stable rather than generating them per request, and warm critical ones after a deploy.
Extraction pipelines lose their retry loops and most of their parsing code.
Smaller, cheaper models become usable for extraction, because format adherence was often the reason to pay for a larger one. Re-run your eval set on a smaller model.
Agent tool calls fail less. Combine strict tool use with argument validation in the tool itself, because a well-formed argument can still be a wrong one.
Once a month: what changed in tech that month and what it means for you, from cloud and AI to security and the tools you build with. Every item links to its source. No sales pitch, unsubscribe in one click.
Book a free 30-minute call about it. Bring your setup and your bill; you'll leave knowing what we'd do first and roughly what it costs, whether or not you hire us.
Or keep reading
Claude Sonnet 4 Retired From Bedrock, and Every Other Model Retiring Before April
Claude Sonnet 4 reached end of life on Amazon Bedrock on October 14, 2026, and requests to it now fail. Claude 3.x ends in more Regions on November 13, Opus 4.1 on January 8, the Llama, Mistral and DeepSeek batch on March 30, and Sonnet 4.5 on April 8. How to find what you call, the gotchas that bite, and how to move without a regression.
Read
Amazon Bedrock Managed Agents Powered by OpenAI: What It Is and When to Use It (Preview)
Bedrock now runs OpenAI's agent harness and frontier models as a managed service inside AWS: IAM permissions, CloudTrail audit, data kept in AWS, with AgentCore Runtime as the default compute. How it relates to AgentCore, the OpenAI Agents SDK and Claude-based agents, and what preview means for production plans.
Read
As of October 25, 2026. Provided as is, for general information, and accurate to the best of our knowledge on the date shown. Not legal, compliance, tax, financial or other professional advice. Prices, limits and vendor services change; check current vendor documentation before relying on anything here. Terms of use.