Anthropic has released Claude Opus 4.6, positioning it as the company’s most capable model yet, with a focus on agentic workflows, large-context reasoning, and tighter integration into professional tools.

A key addition is a new “agent teams” feature in Claude Code, which allows multiple AI agents to collaborate on a single project simultaneously, dividing work across subtasks instead of executing steps sequentially. This marks a clear push toward more autonomous, parallelized software and research workflows.

Opus 4.6 also brings a 1-million-token context window to Anthropic’s Opus tier for the first time, matching what Sonnet already offers for heavy document analysis and large codebases. The expanded context enables sustained reasoning across long-running tasks without aggressive summarization.

New Excel and PowerPoint sidebars embed Claude directly inside Microsoft Office apps, allowing the model to read existing templates and generate spreadsheets or slide decks natively, eliminating the need to copy content between tools.

On benchmarks, Opus 4.6 topped most agentic evaluations and showed a significant jump on ARC-AGI-2, reaching close to 70%, although OpenAI’s Codex 5.3 reclaimed some agentic coding highs shortly after release.

Why it matters: With both OpenAI and Anthropic releasing major agentic upgrades within hours of each other, the pace of frontier model improvement is accelerating. Multi-agent collaboration, massive context windows, and deep workflow integration suggest that AI systems are increasingly being designed not just to assist humans, but to operate as coordinated digital teammates.