The model launch paired capability claims with tools for managing longer tasks and adjusting reasoning effort.
What happened
Anthropic introduced a one-million-token context window in beta, adaptive thinking and API context compaction. The announcement also added experimental agent teams in Claude Code and expanded office-product integrations.
Why it matters
Longer tasks require a model to preserve relevant information, use tools and recover from mistakes. A large context limit alone does not show that every detail will be used correctly.
The bigger picture
Our take. Compare finished work and review time, not only headline benchmark positions. Anthropic's evaluations and early-user reports are useful starting points, but the operating environment and effort setting can change results.
Source checked November 18, 2026. Company claims are attributed; paragraphs marked “Our take” are editorial interpretation.