Anthropic Claude Opus 4.7 Model Released on Amazon Bedrock

Anthropic's Claude Opus 4.7 has been added to Amazon Bedrock. This model improves performance across coding, long-running agents, and professional work overall, and runs on Bedrock's next-generation inference engine, providing infrastructure suited for enterprise workloads.

The new inference engine features scheduling and scaling logic that dynamically allocates capacity per request. As a result, it improves availability for steady-state workload and can also accommodate services that scale rapidly, while maintaining zero operator access, keeping customer prompts and responses invisible to both AWS and Anthropic operators.

The performance improvements are organized around four axes.

• Agentic coding: Shows stronger performance than Opus 4.6 in long-horizon autonomous tasks, systems engineering, and complex code reasoning. According to Anthropic, it recorded SWE-bench Pro 64.3%, SWE-bench Verified 87.6%, and Terminal-Bench 2.0 69.4%.
• Knowledge work: Improved in document writing, financial analysis, and multi-step research workflows. It better interprets ambiguous requests, explicitly states its own assumptions, and self-verifies results, raising the quality of the first pass. It achieved Finance Agent v1.1 64.4%.
• Long-running tasks: Continues work more stably across the 1M token context window, resolving ambiguity and re-verifying outputs.
• Vision: High-resolution image support has been added, improving accuracy on detail-critical inputs such as charts, dense documents, and screen UIs.

However, since this is an upgrade from Opus 4.6, prompting changes and harness adjustments may be needed for optimal performance. AWS advises referring to Anthropic's prompting guide.

Usage methods are also presented. You can test it immediately by selecting Claude Opus 4.7 in the Playground on the Bedrock console, connect to bedrock-runtime via the Anthropic Messages API, or call it through the Invoke and Converse API of the AWS CLI and SDK. The examples show how to generate messages using AnthropicBedrockMantle from the anthropic[bedrock] SDK, or perform the same request via the AWS CLI invoke-model command.

Additionally, Adaptive thinking allows the thinking token budget to be dynamically allocated according to the complexity of the request. Bedrock's Converse API is suited for multi-turn conversation and Guardrails integration, while the Invoke API is advantageous for direct calls and low-level control.

In terms of capacity, the new inference engine is designed to quickly prepare and provide capacity across many models. Under high load, requests are not rejected but are queued, with 10,000 RPM immediately available per account per region, and additional capacity can be requested.

It is currently available in the US East (N. Virginia), Asia Pacific (Tokyo), Europe (Ireland), and Europe (Stockholm) regions, and AWS advises checking details via the Claude by Anthropic in Amazon Bedrock page and the Bedrock pricing page.