In Aleph Alpha's benchmark study of 967 hand-picked taboo topics, Western comparison models Claude Sonnet 5 and Mistral Small provided balanced answers 70% and 92% of the time, respectively