Amazon Bedrock Enforces Schema-Accurate JSON for All Claude Models
In short: AWS has rolled out structured outputs across every Anthropic Claude model hosted on Amazon Bedrock, reaching both the standard runtime endpoint and the newer Mantle endpoint built for agentic workloads. Instead of hoping the model returns valid JSON, developers can now lock responses to an exact schema through constrained decoding, which blocks the model from generating tokens that would break the format. A companion feature, strict tool use, applies the same guarantee to function-calling arguments so agent workflows don't fail on malformed inputs.
This summary was generated automatically by AI from AWS Machine Learning's publication. It is our own text, not a copy of the original — facts, figures and quotes belong to the source, linked above and below.
What changed?
- 1Combines two mechanisms: JSON Schema output format for shaping responses, and strict tool use for validating tool-call parameters
- 2Available for every Claude model wherever it runs on Bedrock, covering both the runtime and Mantle endpoints
- 3Supports batch processing and cross-Region inference on the runtime endpoint, plus real-time streaming as typed events
- 4Schemas are checked against a JSON Schema Draft 2020-12 subset and compiled into a cached grammar that speeds up repeated requests for 24 hours
- 5additionalProperties: false is mandatory on every object; some keywords like recursive schemas, external $ref, and min/max limits are accepted but silently ignored
- 6Demonstrated for lead extraction from emails, ticket classification/routing with Claude Haiku 5.5, and a strict-schema weather tool for agents
| Parameter | Before | Now |
|---|---|---|
| Output format | Free-form text requiring custom parsing and retries | Constrained decoding guarantees JSON matching a defined schema |
| Tool calling | Tool parameters not strictly validated against schema | Strict tool use enforces exact schema match for tool inputs |
| Availability | Not specified | All Claude models on Bedrock runtime and Mantle endpoints where the model is offered |
| Batch/cross-region support | Not specified | Supported with batch inference and cross-Region inference on the runtime endpoint |
| Cost impact | Not specified | Fewer output tokens since responses only include schema-defined fields, potentially lowering inference cost |
Why it matters
For teams building automated pipelines or AI agents, unpredictable JSON formatting has long meant extra parsing code, retries, and silent failures. Enforcing the schema during generation rather than validating it afterward removes a common source of broken integrations and can also trim output tokens — and therefore cost — on high-volume tasks like classification or extraction.
What it means for AI agents and contact centers
If you route call transcripts, ticket data or CRM fields through Claude on Bedrock, structured outputs could remove a layer of validation code and reduce failed tool calls in multi-step voice-agent workflows; worth comparing against current parsing/retry logic for reliability and latency, especially under batch or real-time streaming.
🧪 Worth Testing
Testing strict tool use and JSON Schema output could show whether tool-calling failures and malformed responses drop in agentic call-handling flows, directly affecting automation reliability and integration cost.
Structured Outputs & Strict Tool Use for Claude on Bedrock· New
Sources
- AWS Machine LearningOfficialPrimary sourceOriginal article →„Structured outputs for Anthropic Claude models on Amazon Bedrock“20 Oct 2026, 02:32
- Published by source
- 20 Oct 2026, 02:32
- Found by our system
- 20 Oct 2026, 02:43
- Summary generated
- 20 Oct 2026, 02:46
This article was written by AI from the original source. Facts, numbers and prices come from the source; missing values are marked “Not specified”. Legal notice, copyright and privacy