Anthropic launches Claude Sonnet 4.6 with 1M token beta context and stronger coding workflows

On February 17, 2026, Anthropic introduced Claude Sonnet 4.6, describing it as the most capable Sonnet model yet and a full upgrade across coding, computer use, long-context reasoning, agent planning, knowledge work, and design. The company is also rolling out a 1M token context window in beta, which materially changes the kind of repository-scale or document-heavy tasks a mid-tier model can handle in one pass.

The commercial framing matters as much as the model update. Anthropic says Claude Sonnet 4.6 is now the default model in claude.ai and Claude Cowork for Free and Pro users, while API pricing stays at $3/$15 per million tokens, unchanged from Sonnet 4.5. That combination suggests Anthropic is trying to move more advanced performance out of a premium tier and into a broadly used default model, rather than keeping higher-end reasoning locked behind an Opus-class product.

According to Anthropic, early users preferred Sonnet 4.6 to Sonnet 4.5 by a wide margin, and Claude Code testers preferred it roughly 70% of the time. The company also says users preferred Sonnet 4.6 to Claude Opus 4.5 59% of the time, citing better instruction following, less overengineering, fewer hallucinations, and fewer false claims of success. Those claims come from Anthropic's own evaluations, so outside validation still matters, but the directional signal is clear: the company wants Sonnet to be the practical everyday workhorse for developers and knowledge workers.

One of the more important product angles is computer use. Anthropic says Sonnet 4.6 shows a major jump on OSWorld and is starting to display human-level performance on tasks such as navigating complex spreadsheets and completing multi-step web forms across browser tabs. If that holds up in production environments, it strengthens the case for model-driven automation on legacy enterprise software that lacks APIs or clean integration points.

Anthropic also emphasizes safety. The company says Sonnet 4.6 performed as safely as, or safer than, recent Claude models in its evaluations, and it highlighted improved resistance to prompt injection in computer-use settings. Combined with the longer context window, lower hallucination rate, and unchanged pricing, the release looks less like a flashy benchmark update and more like a deliberate attempt to make agentic, high-context, code-aware AI cheaper to deploy at scale.