--- title: How Claude's values vary by model and language | daily.dev description: Anthropic analyzed 300,000 real Claude.ai conversations to measure how the values Claude expresses vary across models and languages. Using dimensionality... canonical: https://daily.dev/posts/how-claude-s-values-vary-by-model-and-language-cpcsqkgdv twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: How Claude's values vary by model and language | daily.dev og:description: Anthropic analyzed 300,000 real Claude.ai conversations to measure how the values Claude expresses vary across models and languages. Using dimensionality... og:url: https://daily.dev/posts/how-claude-s-values-vary-by-model-and-language-cpcsqkgdv og:image: https://api.daily.dev/og/posts/cpCSqkGdV.png og:image:alt: How Claude's values vary by model and language og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # How Claude's values vary by model and language **[Hacker News](https://daily.dev/sources/hn)** · 17 min read · 1 upvotes · 0 comments ## Summary Anthropic analyzed 300,000 real Claude.ai conversations to measure how the values Claude expresses vary across models and languages. Using dimensionality reduction on 3,307 identified values, they derived four interpretable axes: Deference vs. Caution, Warmth vs. Rigor, Depth vs. Brevity, and Candor vs. Execution. These four axes capture 15% of value variation after controlling for task, topic, and user values. Key findings: Sonnet 4.6 leans toward warmth and deference while Opus 4.7 leans toward rigor and caution, matching public perceptions. Across languages, Claude expresses the most warmth in Arabic and Hindi and the most rigor in English and Russian. The research opens questions about whether language-based variation is desirable, how training data composition drives it, and whether value profiling can become part of model evaluation and monitoring. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### How do the values expressed by Claude Sonnet 4.6 and Opus 4.7 differ? Sonnet 4.6 leans toward deference and warmth, frequently affirming users' ideas and using humor and playfulness, while Opus 4.7 leans toward caution and rigor, often warning of risks unprompted and candidly critiquing the user's work. Opus 4.7 also shows more depth and candor about its own uncertainty, while Opus 4.6 favors brevity and staying within the scope of a request. _Comparing model personalities before picking one for a workflow is easier with these behavioral differences surfaced on daily.dev._ ### Does Claude behave differently depending on what language you use? Yes, Claude's expressed values shift meaningfully by language. It leans toward warmth most in Hindi and Arabic, and toward rigor most in English and Russian; it shows the most deference in Arabic and the most caution in English; and it leans toward candor in Dutch versus execution in Indonesian, based on analysis of conversations across the top 20 languages on Claude.ai. _Developers building multilingual AI products can track findings like this on daily.dev before shipping across locales._ ### How many distinct values has Anthropic identified in Claude's conversational responses? Anthropic identified more than 3,000 distinct values expressed by Claude across 700,000 anonymized Claude.ai conversations in earlier research, later manually clustering them into 339 high-level values for a follow-up study analyzing over 300,000 conversations across three models and 20 languages. _Anyone evaluating LLM behavior at scale can follow this kind of methodology work on daily.dev._ ## Similar posts on daily.dev - [Claude 4.5 Opus' Soul Document](https://daily.dev/posts/claude-4-5-opus-soul-document-vu6fsqiqa) · Hacker News · 2 upvotes · 0 comments - [Anthropic’s Claude Sonnet 5 system card says more about the future of AI than its benchmarks do](https://daily.dev/posts/anthropic-s-claude-sonnet-5-system-card-says-more-about-the-future-of-ai-than-its-benchmarks-do-wvrqbf5ch) · The New Stack · 0 upvotes · 0 comments - [Four points separate software engineers from everyone else using AI](https://daily.dev/posts/four-points-separate-software-engineers-from-everyone-else-using-ai-dtm7ejxmq) · Medium · 1 upvotes · 0 comments - [Claude Model Selection Guide](https://daily.dev/posts/claude-model-selection-guide-pjwnkk9or) · SitePoint · 1 upvotes · 0 comments --- Tags: [#llm](https://daily.dev/tags/llm), [#claude](https://daily.dev/tags/claude), [#anthropic](https://daily.dev/tags/anthropic), [#ai-safety](https://daily.dev/tags/ai-safety), [#ai-governance](https://daily.dev/tags/ai-governance) [View this post on daily.dev](https://daily.dev/posts/how-claude-s-values-vary-by-model-and-language-cpcsqkgdv) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"How Claude's values vary by model and language","url":"https://daily.dev/posts/how-claude-s-values-vary-by-model-and-language-cpcsqkgdv","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/how-claude-s-values-vary-by-model-and-language-cpcsqkgdv"},"datePublished":"2026-07-15T13:22:42.889Z","dateModified":"2026-09-13T13:31:49.586Z","description":"Anthropic analyzed 300,000 real Claude.ai conversations to measure how the values Claude expresses vary across models and languages. Using dimensionality...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/d6f299d2b0868305bb494fd6394e8cf9?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/d6f299d2b0868305bb494fd6394e8cf9?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Hacker News","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Hacker News","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/hn","url":"https://daily.dev/sources/hn"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/how-claude-s-values-vary-by-model-and-language-cpcsqkgdv","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"llm,claude,anthropic,ai-safety,ai-governance","timeRequired":"PT17M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Hacker News","item":"https://daily.dev/sources/hn"},{"@type":"ListItem","position":3,"name":"How Claude's values vary by model and language"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/how-claude-s-values-vary-by-model-and-language-cpcsqkgdv#faq","mainEntity":[{"@type":"Question","name":"How do the values expressed by Claude Sonnet 4.6 and Opus 4.7 differ?","acceptedAnswer":{"@type":"Answer","text":"Sonnet 4.6 leans toward deference and warmth, frequently affirming users' ideas and using humor and playfulness, while Opus 4.7 leans toward caution and rigor, often warning of risks unprompted and candidly critiquing the user's work. Opus 4.7 also shows more depth and candor about its own uncertainty, while Opus 4.6 favors brevity and staying within the scope of a request. Comparing model personalities before picking one for a workflow is easier with these behavioral differences surfaced on daily.dev."}},{"@type":"Question","name":"Does Claude behave differently depending on what language you use?","acceptedAnswer":{"@type":"Answer","text":"Yes, Claude's expressed values shift meaningfully by language. It leans toward warmth most in Hindi and Arabic, and toward rigor most in English and Russian; it shows the most deference in Arabic and the most caution in English; and it leans toward candor in Dutch versus execution in Indonesian, based on analysis of conversations across the top 20 languages on Claude.ai. Developers building multilingual AI products can track findings like this on daily.dev before shipping across locales."}},{"@type":"Question","name":"How many distinct values has Anthropic identified in Claude's conversational responses?","acceptedAnswer":{"@type":"Answer","text":"Anthropic identified more than 3,000 distinct values expressed by Claude across 700,000 anonymized Claude.ai conversations in earlier research, later manually clustering them into 339 high-level values for a follow-up study analyzing over 300,000 conversations across three models and 20 languages. Anyone evaluating LLM behavior at scale can follow this kind of methodology work on daily.dev."}}]} ```