Claude's Varying Tone Across Languages
xiaohu · x · 2026-07-14
Anthropic analyzed 300,000 conversations and found that Claude exhibits different value tendencies based on the input language.
- English: Most cautious and in-depth
- Russian: Most strict, challenging assumptions, correcting details, and demanding evidence
- Hindi: Warmest
- Dutch: Most candid, even willing to admit mistakes
- Indonesian: Most execution-oriented
- Arabic: Most compliant and concise, while also being warm
Compared to the overall average across all conversations, Chinese exhibits a very minor overall shift. The three most prominent behaviors in Chinese conversations are:
- Pointing out competing considerations
- Refuting flawed assumptions
- Offering comfort without judgment
The article concludes that Chinese Claude is neither overly sycophantic nor coldly detached, acting more like an advisor who gently reminds you that "things aren't that simple" while remaining comforting.
Related event: Anthropic Maps How Claude's Values Shift Across Models and Languages(21 posts)→
More from Models
- GPT-6.1 Terra cuts factual error rate from 22.3% to 4.3%, an ~81% drop — haider1 · 2026-10-27
- GPT-6.1 Terra benchmarks: strong score under $1, but Claude Opus 5.5 still leads — 141_1337 · 2026-10-27
- ChatGPT 6 Terra benchmark charts surface, showing score-cost curves — 141_1337 · 2026-10-27
- GPT-6.1 Terra lands: early users praise speed, value and subagent performance — BenBajarin · 2026-10-27
- Leaked GPT-6.1 Terra benchmarks: cheap but 'boring', still far behind Astra and Claude — Angaisb_ · 2026-10-27
- Artificial Analysis Ships Intelligence Index v4.3.2 With GPT-6.1 Terra Comparison — ArtificialAnlys · 2026-10-27