Claude 2.1 is an iteration of Anthropic’s large language model, released in November 2023 to improve accuracy and processing capacity for enterprise applications. It succeeded Claude 2.0, focusing on increased reliability and more sophisticated developer controls.
The model's primary upgrade is a 200,000 token context window, capable of handling approximately 150,000 words in a single prompt. This capacity allows users to upload and query extensive datasets, such as technical manuals, financial reports, or entire software codebases.
Anthropic implemented safety and accuracy improvements resulting in a twofold reduction in false claims compared to its predecessor. Claude 2.1 is designed to be more honest by admitting uncertainty instead of generating hallucinations. It also introduced beta tool use via API, which allows the model to connect to external functions, search web sources, and interact with private knowledge bases.
Additional features include support for system prompts, which provide a way for developers to give persistent instructions that define the model's behavior, persona, and response format.