Xinming Tu analysis shows Claude Opus 5.5 trained on protein sequences

xinmingtu.github.io7 min read

An analysis by researcher Xinming Tu demonstrates that Anthropic's Claude Opus 5.5 model was trained on scale protein sequences. Testing on obscure bacterial proteins with no solved structures or paper mentions revealed that Opus 5.5 predicts subsequent amino acids significantly better than shuffled controls, matching the performance of specialized open-weight protein language models like ProtGPT3-10B. Open article →