Claude Opus 4.6

Source: Anthropic Date: 2026-02-05 URL: https://www.anthropic.com/news/claude-opus-4-6

Summary

Anthropic launched Claude Opus 4.6 with a 1M token context window (beta). Claims: highest on Terminal-Bench 2.0 and Humanity’s Last Exam; +144 Elo points over GPT-5.2 on GDPval-AA. New product features: agent teams in Claude Code, context compaction for longer tasks, adaptive thinking, effort controls. Claude in Excel upgraded; Claude in PowerPoint launches in research preview. Pricing: $5/$25/M tokens standard; $10/$37.50/M for context >200K tokens. 20+ named partners including Notion, GitHub, Figma, and Shopify.

Implications

• Claude model cadence / Claude Code thread. Opus 4.6 is the model running this conversation — it’s Anthropic’s current flagship as of February 2026. The 1M token context is the headline, but agent teams in Claude Code is the operationally significant feature.
• 1M context (beta). 1M tokens covers an entire large codebase or document corpus in a single context. The beta status matters — 1M context at full production reliability is a harder engineering problem than 200K.
• +144 Elo over GPT-5.2 on GDPval-AA. Direct GPT comparison in Anthropic’s own announcement is a new move — they’re citing a head-to-head Elo comparison now. “GDPval-AA” is an agent benchmark worth tracking as it may become an industry-standard eval.
• Premium context pricing. $10/$37.50 for >200K context creates a significant price step — enterprise customers using the full 1M window pay 2x the base rate. This is the monetization mechanism for the capability.
• PowerPoint research preview. Claude in PowerPoint alongside Claude in Excel signals Anthropic is building a Microsoft Office suite integration layer — potentially competing with Microsoft Copilot inside Microsoft’s own products.
• Watch: when 1M context exits beta; GPT-5.x response to the Elo comparison; PowerPoint integration GA timeline; agent teams adoption in Claude Code enterprise customers.