Introducing Claude Opus 4.5 in Microsoft Foundry

We’re at a real inflection point in the AI landscape, a threshold where models move from useful assistants to genuine collaborators. Models that understand the objective, factor in constraints, and execute complex multi-tool workflows. Models that not only support processes, but help restructure them for reliability, scale, and operational efficiency. Anthropic’s newest model, Claude Opus 4.5, embodies that shift. Today, we are excited to share that Opus 4.5 is now available in public preview in Microsoft Foundry, GitHub Copilot paid plans, and Microsoft Copilot Studio. Building on the Microsoft Ignite announcement of our expanded partnership with Anthropic, Microsoft Foundry delivers its commitment to giving Azure customers immediate access to the widest selection of advanced and frontier AI models of any cloud. Foundry empowers developers to accelerate innovation with an integrated, interoperable, and secure AI platform that enables seamless deployment, integration, and scaling for AI apps and agents.

We’re excited to use Anthropic Claude models from Microsoft Foundry. Having Claude’s advanced reasoning alongside GPT models in one platform gives us flexibility to build scalable, enterprise-grade workflows that move far beyond prototypes. —Michele Catasta, President, Replit

Opus 4.5 for real work

Opus 4.5 sets a new bar for coding, agentic workflows, and enterprise productivity: outperforming Sonnet 4.5 and Opus 4.1, at a more accessible price point. Its versatility across software engineering, complex reasoning, tool use, and vision unlocks new opportunities for organizations to modernize systems, automate critical workstreams, and deliver ROI faster. By prioritizing rapid integration of the latest models, Foundry allows Azure customers to stay ahead of the curve and maximize the impact of their agentic AI systems; all while maintaining centralized governance, security, and observability at scale.

1. Built for production engineering and agentic capabilities

According to Anthropic, Opus 4.5 delivers state-of-the-art performance on industry standard software engineering benchmarks, including new highs on SWE-bench (80.9%). Early testers consistently describe the model as being able to interpret ambiguous requirements, reason over architectural tradeoffs, and identify fixes for issues that span multiple systems. Opus 4.5 accelerates engineering velocity by completing multi-day development work in hours with:

• Improved multilingual coding performance
• More efficient code generation
• Stronger test coverage
• Cleaner architectural and refactoring choices

Capability / Benchmark	Claude Opus 4.5	Claude Sonnet 4.5	Claude Opus 4.1	Gemini 3 Pro
Agentic coding (SWE-bench Verified)	80.90%	77.20%	74.50%	76.20%
Agentic terminal coding (Terminal-bench 2.0)	59.30%	50.00%	46.50%	54.20%
Agentic tool use — Retail (t2-bench)	88.90%	86.20%	86.80%	85.30%
Agentic tool use — Telecom (t2-bench)	98.20%	98.00%	71.50%	98.00%
Scaled tool use (MCP Atlas)	62.30%	43.80%	40.90%	_
Computer use (OSWorld)	66.30%	61.40%	44.40%	_
Novel problem solving (ARC-AGI-2 Verified)	37.60%	13.60%	_	31.10%
Graduate-level reasoning (GPQA Diamond)	87.00%	83.40%	81.00%	91.90%
Visual reasoning (MMMU validation)	80.70%	77.80%	77.10%	_
Multilingual Q&A (MMLU)	90.80%	89.10%	89.50%	91.80%

Claude Opus 4.5 benchmark results from Anthropic Opus 4.5 is also one of the strongest tool-using models available today, capable of powering agents that work seamlessly across hundreds of tools. Developers gain access to several important upgrades:

• Programmatic Tool Calling: Execute tools directly in Python for more efficient, deterministic workflows.
• Tool Search: Dynamically discover tools from large libraries without using up space in the context window.
• Tool Use Examples: More accurate tool calling for complex tool schemas.

Together, these capabilities enable sophisticated agents across cybersecurity, full-stack software engineering, financial modeling, and other workflows requiring multiple tool interactions. Opus 4.5 shows strong, real-world intelligence applying these tools creatively within constraints. In testing, the model successfully navigated complex policy environments, such as airline change rules, chaining upgrades, downgrades, cancellations, and rebookings to optimize outcomes. This kind of adaptive, constraint-aware problem-solving reflects a meaningful step forward in what agentic AI systems can accomplish.

Manus deeply utilizes Anthropic’s Claude models because of their strong capabilities in coding and long-horizon task planning, together with their prowess to handle agentic tasks. We are very excited to be using them now on Microsoft Foundry! —Tao Zhang, Co-founder & Chief Product Officer, Manus AI

2. Improved developer experience on Foundry

Opus 4.5 paired with new developer capabilities offered on Foundry is designed to help teams build more effective and efficient agentic systems:

• Effort Parameter (Beta): Control how much computational effort Claude allocates across thinking, tool calls, and responses to balance performance with latency and cost for your specific use cases.
• Compaction Control: Handle long-running agentic tasks more effectively with new SDK helpers that manage context efficiently over extended interactions.

These enhancements provide greater predictability and operational control for enterprise workloads.

3. Enhanced office productivity and computer use

Opus 4.5 also doubles down as Anthropic’s best vision model, unlocking workflows that depend on complex visual interpretation and multi-step navigation. Computer use performance has improved significantly, enabling more reliable automation of desktop tasks. For knowledge workers, the model delivers a step-change improvement in powering agents that create spreadsheets, presentations, and documents. It produces work with consistency, professional polish, and genuine domain awareness making it a fit for finance, legal, and other precision-critical verticals. The model better leverages memory to maintain context and consistency across files throughout sprawling professional projects.

4. Safety and security

According to Anthropic, Opus 4.5 also delivers meaningful improvements in safety and security. The model shows a reduced rate of misaligned responses, stronger robustness against prompt-injection attacks, and more reliable behavior across complex tasks. These improvements align with Microsoft’s commitment to providing enterprise customers with models that meet high bars for safety, governance, and operational integrity

Use cases

Opus 4.5 serves the following use cases

• Software development: Deploy agents that handle complex, multi-system development tasks with minimal supervision.
• Financial analysis: Connect insights across regulatory filings, market reports, and internal data for sophisticated predictive modeling and proactive compliance monitoring.
• Cybersecurity: Correlate logs, vulnerability databases, and threat intelligence for professional-grade threat detection and automated incident response.
• Enterprise operations: Manage sophisticated workflows requiring coordination across multiple tools, systems, and information sources.

Pricing and availability

Opus 4.5 delivers frontier performance and sets a new standard for a variety of use cases at one third the price of previous Opus-class models.

Model	Offer type	Deployment type	Regions	Price (1M tokens)	Availability
Claude Opus 4.5	Serverless Pay-go	Global Standard	East US2, Sweden Central	Input- $5 Output- $25	November 24, 2025 (public preview)

Get started today

Claude Opus 4.5 is available now in Original Postroducts/ai-foundry/" shape="rect">Microsoft Foundry and coming soon in Visual Studio Code via the Foundry extension. Visit the Foundry portal to begin building with Opus 4.5. The post Introducing Claude Opus 4.5 in Microsoft Foundry appeared first on Microsoft Azure Blog.