Anthropic’s Claude Opus 4.5 Debuts on Amazon Bedrock

Anthropic’s cutting-edge foundation model, Claude Opus 4.5, is now available on Amazon Bedrock, offering a significant leap forward in AI capabilities for developers and enterprises. This advanced model outperforms its predecessors, Sonnet 4.5 and Opus 4.1, delivering Opus-level intelligence at a third of the cost, priced at $5 per million input tokens and $25 per million output tokens. Opus 4.5 is specifically engineered for building sophisticated AI agents that can reason, plan, and execute complex tasks with minimal human oversight.

Key features highlight its prowess across various domains. In software engineering, it achieves an impressive 80.9% on SWE-bench Verified, capable of transforming multi-day development projects into hours. It boasts improved multilingual coding, more efficient code generation, better test coverage, and cleaner architectural choices. For office productivity, Opus 4.5 excels at end-to-end project management, creating professional PowerPoint, Excel, and Word documents, including advanced features like document redlining. It also generates higher quality React and HTML artifacts, maintaining crucial consistency and accuracy for finance and other precision-sensitive industries.

As Anthropic’s best vision model, Opus 4.5 scores 80.7% on MMMU, enabling complex visual interpretation and multi-step navigation for tasks like analyzing design mockups or automating browser interactions. For agent developers, it introduces two pivotal improvements: the `tool search` tool allows agents to dynamically discover and load only necessary tools from large libraries, saving tokens and preventing schema confusion. Additionally, `tool use examples` can be provided in tool definitions to enhance accuracy for complex schemas.

Integrated with Amazon Bedrock AgentCore, Opus 4.5 leverages enterprise-grade security and scalability. AgentCore provides persistent memory, a Tool Gateway to convert APIs into agent-compatible tools, and robust observability via CloudWatch for monitoring token usage, latency, and errors. A beta `effort` parameter allows developers to control token spending (high, medium, low) for optimal quality or cost-efficiency. This combination empowers the creation of production-ready agents for long-running workflows in software development, enterprise operations, financial analysis, and cybersecurity.

(Source: https://aws.amazon.com/blogs/machine-learning/claude-opus-4-5-now-in-amazon-bedrock/)