Anthropic has introduced Sonnet 4.6, an updated version of its midsized Sonnet model. This release aligns with the company’s four-month update cycle.
The company highlighted advancements in coding, instruction-following, and computer use capabilities for Sonnet 4.6. This model will serve as the default for users on Anthropic’s Free and Pro plans.
Sonnet 4.6, in its beta release, features a 1 million-token context window. This capacity is double the largest previously available for Sonnet. Anthropic stated this window can accommodate “entire codebases, lengthy contracts, or dozens of research papers in a single request.”
This launch follows the introduction of Opus 4.6 two weeks prior. An updated Haiku model is anticipated in the near future.
Sonnet 4.6 recorded new benchmark scores:
- OS World for computer use
- SWE-Bench for software engineering
- 60.4% on ARC-AGI-2, a measure of human-specific intelligence skills
The ARC-AGI-2 score positions Sonnet 4.6 above most comparable models. However, it ranks below Opus 4.6, Gemini 3 Deep Think, and a refined variant of GPT 5.2.
