Overview
- Claude Sonnet 4.6 is now live as the default model across claude.ai, Claude Cowork and the Claude API, with availability on major cloud platforms.
- The release introduces a beta 1 million‑token context window designed to handle entire codebases, long contracts and multi‑document workflows in one session.
- Early testing showed developers preferred Sonnet 4.6 over Sonnet 4.5 about 70% of the time and over Opus 4.5 in roughly 59–60% of evaluations, citing better instruction following and fewer hallucinations.
- Computer‑use capabilities advance on OSWorld‑style tasks to near‑human performance in navigating spreadsheets and multi‑step web forms, according to Anthropic and early users.
- Anthropic reports stronger defenses against prompt‑injection and safety performance comparable to Opus 4.6, while noting Opus remains the better choice for the deepest reasoning, large‑scale refactoring and multi‑agent coordination; the free tier also gains file creation, connectors and context compaction with usage limits.