Anthropic debuts Claude Opus 4.6 with multi-agent teams and expanded knowledge work

Anthropic has released Claude Opus 4.6, its latest model that improves coding performance over the prior version. The update allows more careful planning, longer sustained agentic tasks, more reliable operation in large codebases, and stronger code review and debugging to identify errors. The model introduces a 1 million token context window in beta testing. Opus 4.6 applies its capabilities to routine professional tasks such as financial analyses, research, and the use and creation of documents, spreadsheets, and presentations. Within the Cowork platform, the model can handle these activities through autonomous multitasking. Anthropic has also enhanced Claude in Excel and introduced a research preview of Claude in PowerPoint. Claude Code now supports the assembly of agent teams that work together on shared tasks. The API adds compaction to summarize context for extended operations, adaptive thinking that adjusts based on context, and effort controls for managing intelligence, speed, and cost. The model recorded the highest score on Terminal-Bench 2.0 and led on Humanity's Last Exam. It outperformed the next best model by approximately 144 Elo points on GDPval-AA and its predecessor by 190 points. Opus 4.6 also achieved the best results on BrowseComp for locating difficult online information. Anthropic stated that the model shows a safety profile comparable to or better than other frontier models, with low rates of misaligned behavior. Claude Opus 4.6 is accessible today through claude.ai, the API, and major cloud platforms. Developers can access it via the claude-opus-4-6 endpoint with unchanged pricing of 5 dollars per million input tokens and 25 dollars per million output tokens.