Anthropic released Claude Opus 4.6 with gains in coding, planning, and agentic work, plus a one-million-token context window in beta. On GDPval-AA, a benchmark of economically valuable knowledge work across finance, legal, and technical tasks, it led GPT-5.2 by about 144 Elo points, and it topped Terminal-Bench 2.0 for agentic coding and Humanity's Last Exam for reasoning. On the MRCR v2 long-context retrieval test it scored 76%, against 18.5% for Sonnet 4.5. New controls included adaptive thinking, four effort levels (low, medium, high, max), and context compaction for long-running tasks; Claude Code gained agent teams for splitting work across agents, and Claude in PowerPoint arrived as a research preview. Pricing held at $5 and $25 per million tokens, rising to $10 and $37.50 for prompts beyond 200K tokens.
5February 2026
ModelConfirmed
Claude Opus 4.6 adds a million-token context window
Participants
Also mentioned
Not parties to this event, but named in the text above.
Tags
- llm
- launch
- context-window
- agent
Sources
Later developments
17 February 2026 · Development
Claude Sonnet 4.6 followed twelve days later and became the default model on the Free and Pro plans in claude.ai and Claude Cowork, at unchanged Sonnet 4.5 pricing of $3/$15 per million tokens. It scored 72.5% on the OSWorld computer-use benchmark, up from under 15% in late 2024.