Anthropic releases Claude 2.1 with a 200K context window

Background

Claude 2.1 followed Claude 2.0, and Anthropic reported its results against that earlier version.^[1]

What happened

On 21 November 2023 Anthropic released Claude 2.1 through its API and in claude.ai.^[1]

Key details

The context window doubled to 200,000 tokens, roughly 150,000 words or over 500 pages. In claude.ai the full window was reserved for Pro users.^[1]

Anthropic reported a 2x decrease in false statements compared with Claude 2.0.^[1]

A beta tool use feature let Claude call developer-defined functions and APIs. A function here is a piece of code the developer writes, which Claude can ask to run in order to get information or take an action.^[1]

System prompts were introduced to let users set Claude's role and style. A system prompt is a set of instructions given before a conversation starts.^[1]

Anthropic also launched the Workbench, a prompt playground in its console, and updated pricing across models at the same time.^[1]

Impact

The release combined longer inputs and fewer false statements with new ways for developers to connect the model to their own software, such as tool use and system prompts.^[1]

The context window is the amount of text a model can consider at once. Doubling it meant a user could paste in far more material, such as a long report or a set of contracts, and ask questions across all of it in one conversation.

Tool use, sometimes called function calling, is a way for a model to ask software to run an action and return a result. Developers define the functions, and the model decides when one is needed. Because it was a beta feature, its details could still change.

A system prompt is a standing instruction that shapes how the model responds across a whole conversation, such as its role or the format of its answers.