Anthropic’s 13 July 2026 research found that the values expressed in Claude’s answers vary by model and by language. Across 309,815 subjective Claude. ai conversations, Claude leaned furthest towards warmth in Hindi, rigour in Russian, caution and depth in English, deference and brevity in Arabic, candour in Dutch, and execution in Indonesian. These are average patterns in different real conversations, after statistical controls. The study did not send identical prompts in different languages, determine why the differences exist, or measure whether users received more accurate, fair or useful answers.Key facts at a glanceQuestionAnswerWhat was studied?Values expressed in Claude’s responses to subjective tasksHow large was the sample?309,815 Claude. ai conversationsWhen was the data collected?A two-week period in May 2026Which users were included?Claude. ai Free, Pro and Max usersWhich models were included?Claude Sonnet 4.6, Opus 4.6 and Opus 4.7How many languages?The 20 most common languages on Claude. aiWhat stood out in Hindi?Claude leaned furthest towards warmthWhat stood out in English?Claude leaned furthest towards caution and depth; it also leaned towards rigourWhich language leaned furthest towards rigour?RussianDid the study compare identical translated prompts?NoDid it measure answer quality or user outcomes?NoAre the three models obsolete?No. All three remain active, although newer flagship models are availableAbout The AuthorAt heart, I am a storyteller drawn to the watershed moments that bend the technology landscape. I braid narrative with data, humanise statistics, and trace the arc from first spark to world-changing impact. My reportage, features and reviews are witty, sardonic, visual and vivid, using anecdote to illuminate rather than eviscerate.