A new Anthropic study analyzes how Claude’s expressed values vary by model and language. Researchers compressed over 3,000 distinct values from 700,000 anonymized Claude.ai conversations into four key axes: Deference vs. Caution, Warmth vs. Rigor, Depth vs. Brevity, and Candor vs. Execution, which together capture 15% of the variation in Claude's values. Comparing models, Opus 4.6 leans toward deference, rigor, brevity, and execution, while Opus 4.7 leans toward caution, rigor, depth, and candor. Across the top 20 languages on Claude.ai, in English Claude leans toward caution, rigor, depth, and candor, while in Arabic it leans toward deference, warmth, brevity, and execution. The approach may ultimately allow connection of value variation to different training decisions.