Anthropic has introduced Claude Opus 4.5, an updated frontier model now available in public preview through Microsoft Foundry, GitHub Copilot paid plans, and Microsoft Copilot Studio. The release follows Microsoft’s recent expansion of its partnership with Anthropic and supports Foundry’s goal of giving Azure customers fast access to a wide range of advanced models within a secure, production-ready environment.
Opus 4.5 delivers performance improvements across coding, agentic workflows, reasoning, and vision, outperforming previous Anthropic models, including Sonnet 4.5 and Opus 4.1. The model is designed to help enterprises modernize software systems, automate complex operational tasks, and deploy higher-fidelity AI agents.
Foundry’s integration enables Azure users to adopt Opus 4.5 quickly while maintaining centralized governance, security, and observability for enterprise-scale deployments.
Engineering and agentic performance
According to Anthropic, Opus 4.5 sets new benchmarks in software engineering tasks, achieving leading scores such as 80.9% on SWE-bench. Early testing indicates the model can interpret ambiguous requirements, reason across multi-component systems, and identify cross-system fixes.
Key engineering improvements included better multilingual coding, more efficient and concise code generation, stronger test coverage and higher-quality architectural and refactoring decisions
Opus 4.5 also expands tool-use capabilities critical for agentic systems with programmatic tool calling for deterministic Python-based execution, tool search to dynamically locate tools without consuming context and tool use examples for more accurate execution on complex schemas
These enhancements support advanced agent workflows in areas such as cybersecurity, full-stack engineering, finance, and operations.
Developer improvements in Foundry
New Foundry capabilities further support Opus 4.5 adoption including effort parameter (Beta) which controls the model’s computational effort to balance reasoning depth, latency, and cost, and compaction control, which helps manage context during long-running, multi-step agent interactions.
These features aim to provide predictable, controllable behavior for production workloads.
Productivity, vision, and computer-use improvements
Opus 4.5 includes Anthropic’s strongest vision capabilities to date, improving reliability in workflows that require document interpretation, UI navigation, and multi-step desktop automation.
For knowledge-worker scenarios—such as creating financial models, legal documents, and presentations—the model shows improved consistency, formatting accuracy, and multi-file context retention.
Safety and security updates
Anthropic reports reductions in misaligned outputs, stronger robustness against prompt-injection attempts, and improved reliability on complex tasks. These updates align with Microsoft’s enterprise requirements around governance, safety, and operational integrity.