Anthropic released Claude Instant 1.2, a new version of its low-latency model, through its API. Incorporating improvements from Claude 2, it scored 58.7% on the Codex coding evaluation (up from 52.8%) and 86.7% on the GSM8K math benchmark (up from 80.9%), the company said, along with better quote extraction and multilingual performance. Anthropic said the model hallucinated less and was more resistant to jailbreaks.