Anthropic has introduced Sonnet 4.6, the latest version of its mid-sized language model, as part of its four-month update cycle. In a post announcing the release, the company highlighted improvements in coding, instruction following, and computer use.
Sonnet 4.6 will become the default model for users on the Free and Pro plans.
The beta version of Sonnet 4.6 includes a context window of 1 million tokens. This is twice the largest window previously offered for Sonnet models. Anthropic said the expanded capacity can handle entire codebases, long legal contracts, or dozens of research papers within a single request.
The update arrives two weeks after the release of Opus 4.6. An updated Haiku model is also expected in the coming weeks.
Anthropic reported new record benchmark results for Sonnet 4.6. These include OS World for computer use and SWE Bench for software engineering tasks.
The model also achieved a score of 60.4 percent on ARC AGI 2, a benchmark designed to evaluate reasoning skills associated with human intelligence. While this result places Sonnet 4.6 ahead of many comparable models, it remains behind Opus 4.6, Gemini 3 Deep Think, and a refined version of GPT 5.2.