Claude Opus 5.5 comes to Microsoft Foundry for long-running coding and knowledge work

Artificial Intelligence (AI) models are now tackling tasks that go beyond just responding to a single request. They are developing features across a codebase, delving into complicated issues, summarising vast amounts of information, and navigating through intricate business processes. 

As tasks become more complex, having raw intelligence is just one piece of the puzzle. The model also needs to maintain its focus, make wise choices throughout the process, clearly communicate its actions, and deliver results that are easy for people to assess and use. Currently, Claude Opus 5.5 has been launched in Microsoft Foundry, offering Anthropic’s latest and most advanced Opus model for developers and businesses looking to create AI applications and agents. 

Claude Opus 5.5 is tailored for tackling complex, everyday tasks. It builds on Opus 5, improving capabilities in coding, knowledge processing, and enduring tasks, while simplifying the understanding of the model’s actions, findings, and future requirements. Additionally, Claude Opus 5.5 works more efficiently, needing fewer tokens, which means lower costs per token and reduced expenses for cache reads. 

Designed for Time-Intensive Tasks 

Creating a simple function is straightforward. However, crafting a feature that integrates multiple services, tracking down production issues throughout a substantial repository, or conducting a task from investigation to implementation and validation is a more complex endeavour. Claude Opus 5.5 is built to handle these extended workflows. 

For software developers, it efficiently manages long-term coding activities such as developing features within a codebase, debugging, refining, and code reviews. It identifies the root cause before making any changes, verifies its progress, and describes modifications in clear language, enabling engineers to quickly verify and trust the updates. 

This combination is invaluable, especially when developers utilise models in agentic coding environments, where sessions may consist of many steps and last for considerable durations. This advantage extends beyond just software development. 

For professionals working with knowledge, Claude Opus 5.5 can collate data from various sources, manage lengthy documents and spreadsheets, conduct analyses, and aid in crafting outputs like memos, reports, and presentations. In comparison to its predecessor, it delivers results that necessitate less editing before they’re ready to be shared. 

An AI Model that Communicates Like a Colleague 

As agents undertake more independent tasks, a new challenge emerges: keeping humans informed without overwhelming them. 

If an agent performs 50 steps, the human should not have to examine all 50 steps to confirm if the task was completed successfully. Claude Opus 5.5 features enhancements to agentic communication, making it easier to follow long-term work. As it operates, the model can highlight the most relevant information: 

• When user input is necessary 

• The outcomes at the conclusion of a lengthy task 

The aim is straightforward: spend less time deciphering what the model did and more time utilising the results. This is particularly important for enterprise agents, where users must comprehend not just the final outcome but also situations where an agent needs clarification, faces limitations, or reaches a decision point requiring human input. 

Adaptive Thinking  

Claude Opus 5.5 employs adaptive thinking, automatically assessing the required level of reasoning for each task. Instead of toggling reasoning on or off or manually setting a thinking-token budget, developers can adjust effort to direct how much effort the model applies to a request. This allows the model to tailor its reasoning according to the task—whether it’s a simple request or a complex problem needing in-depth analysis. 

For developers creating agents, this can minimise the amount of application logic necessary to determine when and how a model should think. 

Built for Long-Lasting Agent Architectures 

Long-lasting agents bring challenges that go beyond just the intelligence of the model itself. Conversations can exceed context limits, available tools for an agent may shift, and applications may need to compact earlier context while retaining an understanding of the work already done. 

Alongside Claude Opus 5.5, Anthropic is launching beta API features aimed at overcoming these challenges, such as asynchronous compaction, keep-tail compaction, and the ability to switch tools during a conversation while maintaining reasoning and prompt caching. These enhancements can help developers ensure continuity during prolonged tasks without needing to reconstruct the application state each time context or tools change. 

Combined with Microsoft Foundry, developers can integrate Claude Opus 5.5 into comprehensive agent systems that connect models with enterprise data, tools, evaluation methods, and operational workflows.

Enhanced Safety Measures for More Powerful Models 

As model capabilities grow, Anthropic is also enhancing the safety measures applied to Claude Opus 5.5. 

Claude Opus 5.5 is the first Opus model to implement safety classifiers, similar to those introduced with Claude Fable 5.1, across areas like cybersecurity, biology, AI development, and distillation. 

For typical developer, educational, and knowledge work scenarios, users can continue employing the model for tasks such as identifying software vulnerabilities or learning about biological concepts. Certain requests recognized by Anthropic as higher-risk or dual-purpose may be redirected to another Claude model equipped with the necessary safeguards. 

This highlights a growing priority in deploying advanced models: advancing model functionality while implementing appropriate safeguards for the new capabilities they introduce. 

Pricing Structure

Utilise Claude Opus 5.5 in Microsoft Foundry 

Selecting a model is just the first step in bringing AI to life.  Microsoft Foundry provides developers with a comprehensive platform to discover models, develop and assess AI applications and agents, connect them with enterprise data and tools, and manage those systems in a live environment. As models evolve to handle more comprehensive responsibilities, the focus has shifted from Can the model respond to this prompt? to Can I rely on it to progress the work?  Claude Opus 5.5 marks another advancement in this journey: improved performance for complex tasks, more adaptable thinking, and clearer interaction between people and the AI systems assisting them. 

Claude Opus 5.5 is currently accessible in Microsoft Foundry.