Claude Sonnet 4.6 Releases with 1M Token Context and Advanced Coding Skills

After months of rumors, Anthropic has officially released Claude Sonnet 4.6. The company calls the latest one its “most capable Sonnet model to date.” Claude Sonnet 4.6 introduces major improvements across coding, computer use, long-context reasoning, agent planning, and knowledge work.

One of the biggest highlights is that the Sonnet 4.6 model now supports a 1M token context window in beta, which makes it possible to handle entire codebases, lengthy contracts, or multiple research papers in a single request. Anthropic claims that early testers have reported that the model’s coding skills and instruction-following consistency make it a preferred option over its predecessor, with some even favoring it over the Opus 4.5 model launched back in November last year. Speaking of Opus, let’s not forget that Anthropic has also released the Claude Opus 4.6 model earlier this month.

Claude Sonnet 4.6 brings major improvements across the board

One of the standout upgrades in Sonnet 4.6 is its computer use capability. Unlike previous models that required special connectors or APIs, Sonnet 4.6 interacts with software much like a human would. Think of clicking, typing, navigating multiple windows, and handling complex tasks in spreadsheets or multi-step web forms.

OSWorld benchmark results confirm significant progress, with Sonnet 4.6 showing human-level capability on a wide range of real-world office tasks. Early users have noted smoother navigation, fewer errors, and better context retention across applications. Besides that, the latest Sonnet model also excels in extended reasoning and multi-step planning.

In the Vending-Bench Arena, Sonnet 4.6 showcased a strategic approach by investing in capacity early before pivoting sharply to profitability, which apparently outperformed competitors in both planning and execution. Apart from coding and planning, users have also reported that Sonnet 4.6 produces visually improved outputs, improved layouts, and professional-grade design results with fewer iterations.

As usual, Anthropic is keeping safety as a major focus area. The company notes that extensive evaluations suggest Sonnet 4.6 is at least as safe as previous models, with strong prosocial behaviors, minimal hallucinations, and effective resistance to prompt injection attacks.

Pricing and availability

On the developer front, Claude Sonnet 4.6 now supports MCP connectors in Excel, letting users pull external context from tools like S&P Global, PitchBook, and Moody’s seamlessly into spreadsheets. Moreover, Claude Sonnet 4.6 is available as the default model for Free and Pro plan users on Claude.ai and Claude Cowork, with pricing unchanged from Sonnet 4.5, which is $3/$15 per million tokens. Last but not least, developers and professionals can begin using claude-sonnet-4-6 via the API immediately.