Overview
- Claude Sonnet 4.6 introduces a beta 1 million‑token context window that can handle entire codebases, lengthy contracts, and multi‑document workflows in single sessions.
- It becomes the default model for Free and Pro users in Claude.ai and Claude Cowork, with availability across the Claude API, major cloud platforms, and Microsoft Foundry.
- Anthropic reports broad upgrades across coding, long‑context reasoning, agent planning, computer use, knowledge work, and design, with notable gains on OSWorld software‑use benchmarks.
- Safety evaluations show stronger resistance to prompt‑injection attacks compared with Sonnet 4.5, with performance resembling the company’s latest Opus model in these tests.
- Early developer testing preferred Sonnet 4.6 over Sonnet 4.5 about 70% of the time and often over Opus 4.5, while pricing remains the same as the prior Sonnet release.