Anthropic Launches Claude Sonnet 4.6 Delivering Frontier Power at Mid-Tier Pricing

Anthropic has officially released Claude Sonnet 4.6, a major evolution of its mid-tier AI model. This update introduces a massive 1 million token context window in beta and significantly enhanced computer use capabilities. It aims to provide enterprise-grade reasoning without the high costs typically associated with flagship models, maintaining the same pricing as the previous version. Developers report a strong preference for Sonnet 4.6 over its predecessors and even the high-end Claude Opus 4.5. The model excels at following complex instructions and refactoring code without the laziness often found in large language models. Early testing shows a 70% preference rate among coders who cite its ability to consolidate logic rather than duplicating it as a key advantage. Find more on Claude pricing here. A standout feature is the refined ability to interact with standard software interfaces. By simulating mouse clicks and keyboard inputs, the model can navigate spreadsheets and web forms like a human. According to the OSWorld benchmark, this version shows steady gains in automating tasks across real software like Chrome and VS Code. This functionality helps organizations automate systems that lack modern API connectors. In benchmarks like OfficeQA, Sonnet 4.6 matches the performance of the high-end Opus 4.6. It holds its own against rivals such as GPT-5.2 and Gemini 3 Pro, particularly in long-horizon planning and strategic reasoning. Users found it less prone to overengineering while maintaining a character that researchers describe as warm, honest, and occasionally funny. Detailed results are available in the system card. The launch includes new tools for developers, such as adaptive thinking and context compaction. For business users, a new Excel integration allows Claude to pull data directly from platforms like S&P Global and Salesforce using the Model Context Protocol. This bridge between AI and existing workflows marks a significant step for office productivity. Safety researchers have given the model high marks for its robust resistance to prompt injections. Despite its increased agency in computer use, evaluations show it is as safe as or safer than previous iterations. The model is now the default for both free and Pro users on the Claude platform, making advanced intelligence accessible to a wider audience immediately.