Claude 2
Claude 2 launches with a public chat beta site, 100K-token prompts, and coding, math and bar-exam gains over Claude 1.3.
Better coding (HumanEval 71.2% vs 56.0%), math (GSM8k 88.0% vs 85.2%) and law (multiple-choice bar exam 76.5% vs 73.0%) than Claude 1.3. Longer outputs. Anthropic claims 2x better at giving harmless responses in internal red-teaming. Initially US and UK only.
- Date
- Tuesday, 11 July 2023
- Lab
- Anthropic
- Kind
- model
- Access
- closed API
- Price
- API offered at the same price as Claude 1.3 (launch page)
Figures
| Measure | Value | Measured by |
|---|---|---|
| Codex HumanEval (Python) | 71.2% vs 56.0% for Claude 1.3 | company |
| GSM8k | 88.0% vs 85.2% for Claude 1.3 | company |
| Bar exam (multiple choice) | 76.5% vs 73.0% for Claude 1.3 | company |
All numbers are Anthropic-run. Launch page lists general availability in US and UK via API and the claude.ai beta site.
Sources
This record was checked against its sources on 6 October 2026. How we check