Anthropic Unveils Claude Sonnet 4.6, Delivering Near‑Opus Performance And Expanded Long‑Context Capabilities

AI safety and research company Anthropic announced that it has introduced Claude Sonnet 4.6, described as its most capable Sonnet model to date. The release is framed as a full upgrade across coding, computer use, long‑context reasoning, agent planning, knowledge work, and design, with a one‑million‑token context window available in beta. For users on Free and Pro plans, Sonnet 4.6 becomes the default model in claude.ai and Claude Cowork, with pricing unchanged from Sonnet 4.5.

The update is positioned as a step that brings higher‑end performance to a broader audience. Developers testing the model early reported that improvements in consistency, instruction following, and contextual understanding made it preferable not only to Sonnet 4.5 but, in many cases, to Anthropic’s more advanced Opus 4.5 model from late 2025. Tasks that previously required an Opus‑class system—particularly those tied to real‑world office workflows—are now presented as achievable with Sonnet 4.6. The company also highlights a notable jump in computer‑use capabilities, an area where earlier Sonnet models lagged.

Anthropic emphasizes that the model underwent extensive safety evaluations. Internal researchers described Sonnet 4.6 as demonstrating strong safety behaviors and no major signs of high‑stakes misalignment, a point the company uses to reinforce its broader positioning around responsible AI development.

The discussion of computer‑use abilities reflects a broader argument about the value of AI systems that can operate software directly rather than through APIs. Anthropic notes that many organizations rely on legacy tools that cannot be automated easily, and that a model capable of interacting with a computer like a human can reduce the need for custom integrations. 

Benchmarks such as OSWorld, which simulate real software environments, show steady gains across sixteen months of Sonnet development. Early users of Sonnet 4.6 report that the model can now handle tasks such as navigating complex spreadsheets or completing multi‑step web forms at a level approaching human proficiency, even if it still trails expert users. At the same time, the company acknowledges risks such as prompt‑injection attacks and claims improved resistance compared with earlier versions.

Sonnet 4.6 Advances Code Quality, Reasoning, And Tool Use

Beyond computer use, Anthropic reports broad improvements across benchmarks. In Claude Code, users preferred Sonnet 4.6 to Sonnet 4.5 in most tests, citing better context reading, reduced duplication, and more reliable multi‑step execution. Many also favored it over Opus 4.5, describing it as less prone to overengineering and more consistent in following instructions. The expanded context window allows the model to work across entire codebases or large research collections, and Anthropic highlights its performance in the Vending‑Bench Arena simulation, where the model adopted a long‑term investment strategy that outperformed competitors.

The company notes that early customers have seen improvements in areas such as frontend development, financial analysis, and visual design quality. Sonnet 4.6 also arrives with updates across the Claude Developer Platform and API, including adaptive and extended thinking modes, context compaction, improved web‑search processing, and expanded tool‑use capabilities. The model is now available across all Claude plans, including the free tier, and can be accessed through Claude Cowork, Claude Code, the API, and major cloud platforms.

The post Anthropic Unveils Claude Sonnet 4.6, Delivering Near‑Opus Performance And Expanded Long‑Context Capabilities appeared first on Metaverse Post.