Claude Outperforms Expert Panel on 23 of 99 Biology Research Problems That Stumped Scientists

Anthropic / X·medium signal

Anthropic's Science Blog published results of a benchmark where Claude was given 99 problems analyzing real biological data against an expert panel. On 23 problems where the experts were stumped, Claude's most recent model solved them. The result demonstrates frontier models' growing capability in scientific reasoning beyond standard NLP tasks, with implications for AI-assisted research acceleration.