Anthropic releases Claude 2.1

Anthropic released Claude 2.1, an update to its flagship model built around three changes: a context window expanded to 200,000 tokens (roughly 150,000 words), a claimed twofold reduction in false statements compared with Claude 2.0, and beta support for tool use — letting the model call external APIs, search the web or query a calculator as part of a response rather than generating text alone.

Anthropic said accuracy on long-document comprehension tasks improved by around 30% alongside the hallucination reduction, though as with the earlier figure this was the company’s own measurement rather than an independently reproduced benchmark. The 200,000-token window was, at release, larger than the context length OpenAI offered for GPT-4, and the tool-use beta put Claude on a similar footing to the function-calling capability OpenAI had shipped for GPT models weeks earlier.

The release landed the same week OpenAI was consumed by the boardroom crisis that briefly removed and then reinstated Sam Altman, and commentary at the time noted that Anthropic had used the moment to ship a substantial update of its own rather than pause. Claude 2.1 became available immediately through Anthropic’s API and powered claude.ai, with the full 200K window initially reserved for paying Claude Pro subscribers.