Claude 2.1: 200K context and fewer false statements
In one sentence Anthropic ships Claude 2.1: 200K-token context window (about 150,000 words, over 500 pages), a 2x decrease in false statements versus Claude 2.0, tool use in beta, system prompts and the Workbench in the Console.
Two weeks after GPT-4 Turbo (128K), Anthropic brings Claude's context window to 200K tokens, about 150,000 words or over 500 pages: you can load an entire codebase, an S-1 filing or a long book in a single request. On claude.ai the 200K context is reserved for Pro users; the model powers both the free and Pro tiers.
The other stated improvement is honesty: according to Anthropic, Claude 2.1 makes half as many false statements as Claude 2.0 and, when unsure, prefers to say it does not know rather than make something up.
Tool use in beta, system prompts and a Workbench in the Console for testing prompts also arrive. Anthropic also updates its pricing to make it more cost-efficient.
Companies
Anthropic
Tools
Claude 2.1
Tags
Sources