Anthropic's latest flagship model, Claude Opus 4.7, is now hosted on Amazon Bedrock. For engineering teams, the more significant update may be Bedrock's new inference engine powering it. This engine introduces fresh scheduling and scaling logic that dynamically allocates capacity. The design aims to maintain availability for steady-state workloads while accommodating services that need to scale quickly, queuing requests instead of rejecting them during peak demand.

Architecturally, the service maintains zero operator access; prompts and responses are not visible to AWS or Anthropic personnel. The Opus 4.7 model itself shows measured improvements in areas like agentic coding, where it scores 64.3% on SWE-bench Pro, and long-context tasks within its 1-million-token window. It adds higher-resolution image analysis for technical documents and UIs.

Integrating the model requires choosing an API path. Bedrock offers its native Converse API for multi-turn conversations or the lower-level Invoke API. Alternatively, engineers can use the Anthropic Messages API via the `anthropic[bedrock]` SDK, which handles AWS SigV4 authentication. The new engine provides a baseline of 10,000 requests per minute per region, with more available on request.

For production use, note that Opus 4.7 may require prompt engineering adjustments compared to its predecessor. It's currently deployed in regions including US East (N. Virginia), Europe (Ireland), and Asia Pacific (Tokyo).

Source: AWS