Anthropic released Claude Sonnet 4.6 with a 1 million-token context window in beta. The company said the update to its midsized AI model improved coding, instruction-following and computer use compared with earlier Sonnet models.
Sonnet 4.6 became the default model for Free and Pro users in claude.ai and Claude Cowork. It was also available through Claude Code, Anthropic’s API and major cloud platforms, including Amazon Bedrock.
Expanded context window
Anthropic said the larger context window could accommodate entire codebases, lengthy contracts or dozens of research papers in a single request. The company also reported improvements in reasoning across long inputs and planning tasks involving multiple steps.
The launch followed Claude Opus 4.6, which also introduced a 1 million-token context window in beta.
Benchmark results and limitations
Anthropic presented computer-use and software-engineering benchmark results, including OSWorld and SWE-bench, as evidence of improvements over previous Sonnet models.
Reported results for Sonnet 4.6 also included a 60.4% score on ARC-AGI-2, a benchmark designed to assess generalization to unfamiliar problems.
Anthropic cautioned that OSWorld tests a defined set of computer tasks in a controlled environment and does not provide a complete picture of real-world computer use. The company also acknowledged that Sonnet 4.6 still trailed the most skilled humans at using computers.
