Anthropic has unveiled Claude Sonnet 4.6, its latest Sonnet model, marking a substantial upgrade across key AI capabilities. The company reports notable improvements in coding proficiency, computer interaction, long-context reasoning, agent planning, knowledge work, and design.
A Leap in Performance
Sonnet 4.6 delivers performance that Anthropic claims previously required their top-tier Opus models. Early testers, including developers, have shown a strong preference for Sonnet 4.6 over its predecessor, Sonnet 4.5, and even over the November 2025 Claude Opus 4.5. This suggests a significant leap in efficiency and effectiveness for everyday office tasks.
A standout feature is the expanded 1M token context window, currently in beta. This allows the model to ingest and reason over vast amounts of data, such as entire codebases or extensive documentation, in a single prompt. This capability was demonstrated in the Vending-Bench Arena, where Sonnet 4.6 employed a strategic approach to maximize profits over a simulated business lifecycle.
Mastering Computer Interaction
The model also shows marked improvements in its ability to interact with computer systems. Unlike previous AI models that required custom connectors for specialized software, Sonnet 4.6 can navigate and operate applications like a human user would, using a virtual keyboard and mouse. This advancement addresses the challenge of automating tasks within legacy systems.
Anthropic highlights the OSWorld benchmark, which simulates real-world software interactions, as evidence of progress. Sonnet 4.6 demonstrates human-level capability in tasks like complex spreadsheet manipulation and multi-step web form completion. While still not matching expert human users, the rapid progress in this area is notable, making AI more practical for a wider range of work.