Anthropic releases Sonnet 4.6

Anthropic has rolled out a new version of its mid-sized Sonnet model, sticking closely to the company’s roughly 4-month refresh rhythm. In the announcement post, Anthropic highlighted stronger performance in coding, better instruction-following, and improved computer-use capabilities.

Sonnet 4.6 will become the default model for users on the Free and Pro plans.

The beta version of Sonnet 4.6 will ship with a 1–million–token context window, which is double the largest previously offered for Sonnet. Anthropic described this expanded window as “enough to hold entire codebases, lengthy contracts, or dozens of research papers in a single request.”

The update arrives only two weeks after the release of Opus 4.6, and an updated Haiku model is expected to follow in the next few weeks.

Alongside the launch, Anthropic is also pointing to a new batch of record benchmark results, including OS World for computer use and SWE-Bench for software engineering. The headline number, however, may be its 60.4% score on ARC-AGI-2, a test intended to measure capabilities associated with human intelligence. That result places Sonnet 4.6 ahead of most models in its class, though it still sits behind models such as Opus 4.6, Gemini 3 Deep Think, and one refined version of GPT 5.2.