On Thursday, Anthropic unveiled Opus 4.6, the newest iteration of its most advanced model and a key pillar of Claude Code. The previous version, Opus 4.5, launched just last November, but with 4.6 the company is aiming to significantly expand the model’s capabilities and broaden its appeal to a wider range of users and use cases.
The standout upgrade in Opus 4.6 is the introduction of what Anthropic calls “agent teams.” This feature allows complex tasks to be broken down and distributed across multiple agents, each responsible for a specific part of the job. Rather than relying on a single agent working step by step, the work can now be handled in parallel, with agents coordinating directly with one another.
Anthropic describes the experience as similar to managing a highly skilled human team. Scott White, the company’s Head of Product, said that dividing responsibilities in this way enables agents to collaborate simultaneously and complete work more quickly. Agent teams are currently available as a research preview for API users and subscribers.
Opus 4.6 also introduces a significantly expanded context window. The model now supports up to 1 million tokens, allowing it to retain and process far more information within a single session. This brings Opus in line with Anthropic’s Sonnet models (versions 4 and 4.5) and makes it better suited for tasks involving large codebases or extensive documents.
Another notable enhancement is deeper integration with PowerPoint. Claude is now embedded directly into PowerPoint as a side panel, streamlining the presentation workflow. Previously, users could ask Claude to generate a deck, but they had to export the file and open it separately to make edits. With the new integration, presentations can be created and refined directly inside PowerPoint, with Claude offering real-time assistance throughout the process.
