Anthropic has launched an updated iteration of its Sonnet model, maintaining its usual four-month release schedule. The company’s announcement post for the new model highlighted enhanced capabilities in coding, adherence to instructions, and general computer interaction.
Users on both Free and Pro plans will now have Sonnet 4.6 set as their default model.
The Sonnet 4.6 beta introduces a 1-million-token context window, doubling the previous maximum size available for Sonnet. Anthropic stated that this expanded context window is “sufficient to accommodate full codebases, extensive contracts, or numerous research papers within a single query.”
This launch follows closely, just two weeks after the introduction of Opus 4.6, and an updated Haiku model is anticipated in the near future.
The introduction of Sonnet 4.6 is accompanied by several new benchmark highs, notably in OS World for computer operation and SWE-Bench for software development. Its 60.4% achievement on ARC-AGI-2, a test designed to evaluate human-like intelligence, stands out as particularly remarkable. While this score positions Sonnet 4.6 ahead of most similar models, it remains slightly behind top-tier models such as Opus 4.6, Gemini 3 Deep Think, and a specialized variant of GPT 5.2.