Anthropic launches Claude Opus 4.6 with 1M token context and Agent Teams

Anthropic has released Claude Opus 4.6 as its new flagship model, succeeding Opus 4.5, with a focus on improving Claude Code workflows. Opus 4.6 is the first Opus model to offer a one million token context window for working with larger documents and codebases in a single session, and it increases maximum output to 128,000 tokens.

The release also introduces Agent Teams, which let multiple agents split and coordinate on complex tasks in parallel. The feature is available as a research preview in Claude Code and for API subscribers. Anthropic is also adding Adaptive Thinking, where the model decides when deeper reasoning is needed, and Compaction, which summarizes older context as sessions near the context window limit.

On benchmarks, Opus 4.6 is reported at 1606 Elo on GDPval AA, ahead of GPT 5.2 and Opus 4.5, and it also posts strong results on Humanity’s Last Exam, Terminal Bench 2.0, and BrowseComp, although Anthropic notes that these results may not fully reflect real world performance. Anthropic also updated its Excel integration to better handle unstructured data and added a PowerPoint integration that embeds Claude as a side panel for editing presentations.