Claude Sonnet 4.6 brings 1M token power and fewer AI hallucinations

Anthropic has released Claude Sonnet 4.6, the latest version of the company’s mid-range AI model, according to a recent blog post. The update promises significant improvements, especially for those using Claude AI for coding and more advanced tasks.

According to Anthropic, around 70% of developers with early access to Sonnet 4.6 prefer the new model over its predecessor. Sonnet 4.6 is described as more consistent, more adept at following instructions, and less prone to giving mind-blowing answers or falsely claiming tasks have been accomplished – things we all want our LLMs.

Another new feature is a significantly expanded pop-up of up to 1 million tokens (in beta). In practice, this means that the model can handle very large amounts of text in a single query.

Sonnet 4.6 also improved “computer use skills”, meaning it can interact with programs in a more human-like way by clicking, typing and navigating, for example, web browsers and spreadsheets. In the OSWorld benchmark test, the model shows clear improvements compared to previous versions.

Claude Sonnet 4.6 is now the standard model for Free and Pro users of claude.ai and Claude Cowork.