Claude’s personality shifts by model and language: new Anthropic study

Anthropic has released a new study suggesting Claude doesn’t have one fixed personality. The assistant’s tone can shift depending on which Claude model you’re using, and even on the language you use to talk to it. Anthropic says that switching between models or languages can change how Claude sounds, and how it handles judgment calls.

For the study, the company reviewed 309,815 anonymized Claude chats across Claude Sonnet 4.6, Claude Opus 4.6, and Claude Opus 4.7. It focused on prompts involving subjective advice, feedback, and planning, then measured the replies across four dimensions: deference versus caution, warmth versus rigor, depth versus brevity, and candor versus execution.

The differences were pretty clear. Claude Sonnet 4.6 came across as warmer, quicker, and more affirming. Claude Opus 4.7 was more cautious, more rigorous, and deeper in its responses, and it was also more likely to question assumptions or flag risks. Language changed the feel too. Hindi and Arabic read as the warmest, and in some cases more polite or more humorous, while English and Russian leaned more skeptical. English scored highest on caution and depth, while Arabic scored highest on deference and brevity.

That matters if you use Claude for brainstorming, second opinions, travel plans, career advice, or tougher decisions. Switch models or switch languages, and you may get a different tone and a different kind of judgment. Anthropic says it still isn’t clear whether that should be seen as a consistency problem or as a useful form of cultural adaptation.

You can use Claude through Anthropic’s platform, but this public analysis only covers the older models Claude Sonnet 4.6, Claude Opus 4.6, and Claude Opus 4.7. It doesn’t include newer models such as Claude Sonnet 5 or Claude Opus 4.8.