Researchers Say Anthropic's Claude Fable 5.2 Predicts Unstudied Proteins at 54%
Summary
Researchers Xinming Tu and Leo Tianlai Chen report that Anthropic's Claude Fable 5.2 and Opus 5.5 predict the next amino acid in little-studied bacterial proteins, evidence they say points to protein sequence training data. Fable 5.2 scored 54% on real proteins versus 9% on shuffled copies; GPT, Gemini and seven open-weight models stayed near zero.
Key Points
- Tu and Chen report Claude Fable 5.2's gap over shuffled proteins at 45.1 points, ahead of ProGen3-3B's 32.5.
- In 128 protein families with no papers at all, Fable 5.2's advantage holds at 45.9 points, the researchers say.
- Claude Opus 5.5 predicts human and mouse coding DNA but not introns, promoters or intergenic regions, the researchers find.