Anthropic on Thursday released Opus 4.6, the latest version of its most advanced AI model and a key engine behind Claude Code. The update arrives less than a year after Opus 4.5 and is designed to broaden the model’s appeal beyond software developers to a wider range of enterprise and knowledge work use cases.
The most notable addition is the introduction of “agent teams,” a feature that allows multiple AI agents to divide and coordinate work on complex tasks. Instead of a single agent completing steps sequentially, tasks can now be split into parallel segments handled by separate agents that communicate with one another. Anthropic Head of Product Scott White compared the approach to assigning work across a skilled human team, allowing faster execution and clearer ownership of responsibilities. Agent teams are currently available as a research preview to API users and subscribers.
Opus 4.6 also significantly expands its context window, enabling the model to process up to one million tokens per session. This places it on par with Anthropic’s Sonnet models and allows users to work with larger code bases, longer documents, and more complex workflows without losing context. The expanded memory is aimed at professional use cases where continuity and scale are critical.
Deeper Integration Into Workplace Tools
Anthropic has also deepened Claude’s integration into Microsoft PowerPoint. Claude now appears as a side panel within the application, allowing users to create and edit presentations directly without exporting files between tools. Previously, users could generate slides using Claude but had to manually transfer and modify them in PowerPoint. The update positions Opus as a more seamless productivity assistant for business users.
According to White, Opus has evolved beyond its original focus on software development. While Claude Code remains popular among engineers, Anthropic has seen increased adoption by product managers, financial analysts, and other professionals who rely on the model for structured reasoning and complex task execution.
OpenAI Responds With New Codex Model
The release of Opus 4.6 was quickly followed by a response from OpenAI. On Monday, OpenAI launched Codex, an agentic coding tool aimed at developers, and on Thursday also launched a new underlying model, GPT-5.3 Codex. The company said the model significantly expands Codex’s capabilities beyond code writing and review, enabling it to perform a wide range of computer-based tasks typically handled by developers and other professionals.
OpenAI claims GPT-5.3 Codex is 25% faster than its predecessor and capable of building complex applications and games over multi-day workflows. The company also said early versions of the model were used internally to help debug and evaluate the system itself.
The timing of the releases highlights intensifying competition between OpenAI and Anthropic in the emerging market for agentic AI tools. That rivalry has increasingly played out in public, from near-simultaneous product launches to recent sparring over Super Bowl advertisements mocking AI monetization strategies. While the debate over ads sparked online backlash, the rapid cadence of competing releases suggests neither side is laughing much, as the contest for developers, enterprises, and public trust continues to accelerate.