Anthropic Study Shows Claude's Behavior Varies Across Model Versions and Languages

According to research published on Monday, Anthropic analyzed 309,815 anonymized user conversations with Claude to study how the AI expresses values across different contexts. The company distilled more than 3,300 identified values into four behavioral dimensions: deference vs. caution, warmth vs. rigor, depth vs. brevity, and candor vs. execution.

Anthropic found each Claude model exhibited distinct behavioral profiles. Sonnet 4.6 emphasized warmth, deference, and brevity, while Opus 4.7 emphasized rigor, caution, candor, and depth. Claude's behavior also varied by language: Arabic responses were more deferential and warmer, while English and Russian responses were more rigorous and challenging. Dutch responses showed greater candor about uncertainty, whereas Arabic responses tended to be more concise.