Anthropic admits Claude has multiple personalities: It appears especially warm and friendly with Hindi users

Do you think AI responds the same way in every language? Anthropic says no. Its latest study revealed that Claude comes across as warmer and friendlier in Hindi, using more humour, politeness, and words of encouragement than it typically does in English.

You might think that asking Claude the same question in Hindi or English would only change the language of the response. However, a new study by Anthropic suggests it also alters the chatbot's behaviour. According to the company, Claude exhibits different behavioural traits depending on the language users choose. In Hindi, for example, the AI is more likely to respond with humour, politeness, and encouragement than in English.

In its latest research, Anthropic analysed over 309,000 anonymous conversations with Claude -- covering three models (Sonnet 4.6, Opus 4.6, and Opus 4.7) and the 20 most-used languages on the platform -- focusing primarily on queries seeking advice, feedback, and opinions: situations where there is no single correct answer. The company noted that its goal was not to measure accuracy, but to understand how Claude communicates with users.

What were the results? Anthropic found that Claude conveyed greater warmth when responding in Hindi and Arabic. Conversely, its responses in English and Russian were more analytical and focused on rigor.

To facilitate the comparison of results, Anthropic classified Claude's behaviour into four general categories: deference vs. caution, warmth vs. rigor, depth vs. brevity, and candor vs. execution. According to the company, this approach allowed researchers to focus on Claude's behaviour rather than on differences between the questions asked by users.

For users in India, the study's most significant finding lies in what Anthropic calls the "warmth vs. rigor" axis. The company observed that Claude expressed the highest levels of warmth in Hindi and Arabic. In Hindi, the AI tended to use polite language, humour, and a light-hearted tone, while validating users' ideas and work. It also frequently offered unsolicited reassurance, adapted its tone to the user's emotional state, and encouraged people to aim higher. "The greatest variation is observed along the warmth-versus-rigor axis: Claude tends to express values associated with warmth primarily in Arabic and Hindi, and values associated with rigor mainly in English and Russian," Anthropic noted in its blog post.

English, however, presented a different picture. Anthropic found that Claude was more inclined to respond with rigor and caution, questioning assumptions, correcting details without being asked, and backing up its answers with evidence. Overall, the company stated that the most significant differences across languages lay in the degree of warmth or rigor Claude displayed, while its other behavioural traits remained relatively constant.

The study also revealed that Claude's behaviour varied depending on the model used. Sonnet 4.6 was perceived as the warmest; it often employed humour, adapted to the user's tone, and offered non-judgmental reassurance. Opus 4.7, by contrast, appeared more analytical: it tended to question assumptions, explain its reasoning, point out potential risks, and openly acknowledge its own limitations. Opus 4.6 fell somewhere in between, generally sticking to the user's request and getting straight to the point.

Anthropic clarifies that these findings do not imply that Claude holds different beliefs. Rather, the company suggests that the results reflect differences in how the AI responds depending on the language and model. Anthropic added that it is continuing to investigate the causes of these variations but believes the research could help achieve greater consistency in future AI systems.