Anthropic has launched Claude Sonnet 4.6, its most capable mid-tier model to date, now serving as the default model across claude.ai and Claude Cowork.
Sonnet 4.6 introduces a 1 million token context window in beta, enough to hold entire codebases or dozens of research papers in a single request. Early users preferred it over its predecessor roughly 70% of the time, citing better reasoning, fewer hallucinations, and stronger follow-through on multi-step tasks. On OSWorld-Verified, a benchmark for AI computer use, it scored 72.5%, which was up from Sonnet 4.5’s 61.4% and nearly five times the score recorded by Claude 3.5 just sixteen months ago.
How It Compares
On the SWE-bench coding benchmark, Sonnet 4.6 scored 79.6%, putting it firmly in frontier territory. Its computer use score of 72.5% stands well ahead of GPT-5.2’s 38.2% on the same test. Within Anthropic’s lineup, Opus 4.6 still leads on graduate-level reasoning and complex terminal coding, but for most enterprise workloads, the gap has effectively closed, at one-fifth the cost.
The Bottom Line
Performance that previously required a premium Opus-class model is now available at Sonnet pricing, a shift that puts real pressure on competitors whose flagship offerings no longer command an unambiguous advantage. Sonnet 4.6 is available now on claude.ai, Claude Code, Amazon Bedrock, and Azure AI Foundry.
