Anthropic conducted research revealing that its chatbot, Claude, alters its communication style depending on the model version and the language used in the conversation. The study was based on an analysis of 300,000 real-world dialogues.
The Opus 4.7 version most frequently acts as a “helpful critic,” pointing out flaws in the user’s plans, asking clarifying questions, and suggesting a risk assessment. Meanwhile, the Sonnet 4.6 model delivers quick, encouraging responses.
It was also observed that in Hindi and Arabic, Claude comes across as “warmer and friendlier,” whereas in English and Russian, it sounds more serious, meticulous, and skeptical.
Anthropic has yet to reach a consensus on whether these differences should be considered a problem or a natural adaptation to cultural norms.
About the Author
