Claude Instant 1.2
Claude Instant 1.2 folds Claude 2's strengths into the cheap, fast tier, scoring 58.7% on Codex HumanEval and 86.7% on GSM8K.
Upgrade of the fast, lower-priced model over Instant 1.1 (52.8% HumanEval, 80.9% GSM8K) with longer, better-structured answers, stronger quote extraction and multilingual ability, and lower hallucination and jailbreak rates in Anthropic's red-teaming. Available through the API.
- Date
- Wednesday, 9 August 2023
- Lab
- Anthropic
- Kind
- model
- Access
- closed API
Figures
| Measure | Value | Measured by |
|---|---|---|
| Codex HumanEval | 58.7% vs 52.8% for Claude Instant 1.1 | company |
| GSM8K | 86.7% vs 80.9% for Claude Instant 1.1 | company |
Company-run numbers on public benchmarks.
Sources
This record was checked against its sources on 6 October 2026. How we check