Claude 2.1
Claude 2.1 doubles the context window to 200K tokens, halves false statements versus Claude 2.0, and adds system prompts and tool-use beta.
200K tokens (about 150,000 words, 500+ pages), double Claude 2. Anthropic reports a 2x decrease in false statements versus Claude 2.0 and 30% fewer incorrect answers on document-analysis tasks. Adds system prompts and a beta of tool use (calling APIs, databases, calculators).
- Date
- Tuesday, 21 November 2023
- Lab
- Anthropic
- Kind
- model
- Access
- closed API
- Price
- Launch page mentions updated pricing; rates are in a linked PDF not captured
Figures
| Measure | Value | Measured by |
|---|---|---|
| Context window | 200K tokens about 150,000 words | company |
| False statements vs Claude 2.0 | 2x decrease Anthropic internal honesty evals | company |
| Incorrect answers on document tasks | -30% vs Claude 2.0 | company |
At launch the 200K window was reserved for Claude Pro subscribers on claude.ai. Tool use here is a beta that became generally available on 2024-05-30.
Sources
This record was checked against its sources on 6 October 2026. How we check