Sonnet 4.6 Update.

Anthropic has launched Claude Sonnet 4.6, positioning its latest mid-tier model to deliver performance historically reserved for its premium 'Opus' class. The update, which is now the default model for users on claude.ai, focuses on substantial improvements in coding, agentic computer use, and complex reasoning over long documents. The release signals a strategic effort to make high-end AI capabilities more accessible for a wider range of business and development tasks without increasing costs, directly challenging the market's performance-to-price equilibrium.

Available immediately through its API and on major cloud platforms, Sonnet 4.6 maintains the same pricing as its predecessor—starting at $3 per million input tokens and $15 per million output tokens—while incorporating a 1-million-token context window, currently in beta. A key technical advancement is the model's ability to operate software through a virtual mouse and keyboard, a skill measured by the OSWorld-Verified benchmark, showing steady progress in automating tasks within common applications like Chrome and LibreOffice. Anthropic also reports significant safety enhancements, noting Sonnet 4.6 is more resistant to prompt injection attacks than previous versions.

The introduction of Sonnet 4.6 is set to influence how organizations procure and deploy AI systems, offering a viable, lower-cost alternative for workloads that previously required more expensive, frontier models. Early customer reports from firms like Box and Rakuten AI cite double-digit performance increases in enterprise Q&A and improved code generation quality, suggesting the model effectively closes the gap with top-tier competitors for many practical applications. This move could accelerate the adoption of complex agentic workflows and place pressure on other AI providers to deliver similar levels of performance within their own mid-range offerings.

Anthropic's strategy with Sonnet 4.6 is to rapidly commoditize near-frontier performance, shifting the competitive landscape from a focus on peak model intelligence to the scalable and economical deployment of advanced AI capabilities.