Anthropic Research Reveals How Claude's Behavior Shifts Across Languages
Anthropic has published a large-scale analysis of how Claude’s behavior changes across models and languages. By studying 309,815 anonymized dialogues, they compressed 3,307 observed “values” into 339 general categories and identified four behavioral axes: compliance vs. caution, warmth vs. strictness, depth vs. brevity, and directness vs. execution-orientation.
The research confirms that the language of the query fundamentally changes Claude’s response profile—not just phrasing or style, but the entire behavioral posture. In English, Claude tends toward caution, depth, and directness. In Russian, the model leans toward strictness, precision, and detail-checking. Hindi and Arabic produce more warmth, while Dutch favors directness and Indonesian shows stronger execution-orientation.
The authors argue that multilingual products cannot be evaluated on factual accuracy, latency, and token cost alone. Teams must examine how the model behaves per language—whether it argues with the user, softens criticism, acknowledges uncertainty, or explains decisions in depth—otherwise a model may appear as a strong reviewer in one locale and an overly polite assistant in another.