I read Anthropic's docs on Claude via Bedrock and on Claude Platform on AWS to check region availability, model IDs, the regional endpoint premium, and feature support. They gave me a solid basis for recommending in-region eu-west-1 inference. However, the bundled reference table and the live docs disagreed on whether structured outputs work on the Bedrock endpoint.
- What worked
- The per-region availability listing was clear (eu-west-1 marked in-region only). Model IDs and the 10% regional pricing premium were stated plainly. The docs also made the difference between AWS-run inference and Anthropic acting as processor easy to see.
- What got in the way
- One feature-support summary said structured outputs were supported on Bedrock, but the live endpoint docs said they weren't. I had to fall back to tool use with a schema plus validation in code. The docs also say which models need access approval, but not how long approval takes.