Anthropic has released Claude Opus 4.6, an upgraded version of its flagship AI model that can handle longer conversations and coordinate multiple AI agents working simultaneously on complex tasks. The company claims the model outperforms competitors including OpenAI’s GPT-5.2 on several professional benchmarks.
The release introduces a one-million-token context window for the first time in Anthropic’s Opus-class models. This expansion allows the AI to process substantially more information without performance degradation, a persistent problem the industry calls “context rot.” On retrieval tests, Opus 4.6 scored 76 percent compared to 18.5 percent for Sonnet 4.5.
Anthropic positions the new model as particularly suited for enterprise work. The company states Opus 4.6 can perform financial analyses, research tasks, and work with documents, spreadsheets, and presentations. In coding applications, the model can navigate large codebases and sustain longer autonomous workflows.
Agent Teams and reasoning
A new feature called agent teams allows multiple Claude instances to work in parallel on different aspects of a project. One session acts as team lead, coordinating work and assigning tasks, while teammates operate independently in their own context windows. Unlike subagents that report back to a main agent, team members can communicate directly with each other and users can interact with individual teammates.
Agent teams remain experimental and disabled by default. The feature works best for tasks requiring parallel exploration, such as code reviews where different teammates focus on security, performance, and test coverage simultaneously. Anthropic acknowledges the approach adds coordination overhead and uses significantly more tokens than single sessions.
The company provides four effort levels that let developers control how deeply the model reasons through problems. At the default high setting, Opus 4.6 uses extended thinking when useful. Anthropic recommends dialing down to medium for simpler tasks where deeper reasoning adds unnecessary cost and latency.
New API features include adaptive thinking, which allows Claude to decide when deeper reasoning helps, and context compaction, which automatically summarizes older context to enable longer-running tasks without hitting limits. The model supports outputs up to 128,000 tokens.
Safety measures and Enterprise features
Anthropic emphasizes that capability gains did not compromise safety. The company states Opus 4.6 shows low rates of problematic behaviors like deception and sycophancy while achieving the lowest rate of over-refusals among recent Claude models. Anthropic developed six new cybersecurity probes to detect potentially harmful uses of the model’s enhanced capabilities.
The release arrives three days after OpenAI launched a desktop application for its Codex AI coding system. Competition between the companies has intensified as both pursue enterprise customers. Anthropic reported in November that Claude Code reached one billion dollars in run rate revenue six months after becoming generally available.
Survey data from Andreessen Horowitz shows Anthropic’s enterprise adoption has grown from near zero in early 2024 to approximately 40 percent of surveyed companies using it in production by January 2026.
The product expansion includes Claude integration with PowerPoint in research preview, allowing users to create presentations within Microsoft’s application. This places Anthropic’s AI inside a core Microsoft product despite Microsoft’s major stake in OpenAI.
Anthropic also announced improvements to Claude in Excel, enabling the AI to handle multi-step changes, ingest unstructured data, and plan before acting. The company says these updates make Claude more capable for everyday work tasks.
Video
Sources: Anthropic News, Bloomberg, ZDNet, Claude Code Docs, VentureBeat
