The Political Compass places each model on two axes — left/right on economics, and authoritarian/libertarian on social questions. We ran 6 models in 7 languages. The striking part isn’t where any one lands — it’s that they all land in the same corner.
6 models7 languages42 resultscollected 22 Mar 2026
42 / 42
results land in the economically-left, socially-libertarian quadrant — with no exceptions.
Chinese
is the most centrist language; German the most left — the same model shifts with the language you use.
Grok 4
is the one model that breaks from the pack — furthest to the economic right of the field.
1Convergence
Every model, every language, the same quadrant.
Each dot is one model answering in one language. All 42 of them fall in the bottom-left — economically left and socially libertarian. Not a single model, in any language, lands anywhere else.
Claude Sonnet 4.6DeepSeek V3.2Gemini 2.5Grok 4MistralGPT-5.4
Models cluster tightly on the social axis and spread more on economics — which is exactly where the disagreements below show up. Explore every model on the live compass →
2Language shifts the result
Ask the same models in a different language, and the needle moves.
Averaged over all models, the language of the prompt shifts the economic position in a consistent order — German the most left, Chinese the most centrist. The effect is model-dependent: DeepSeek V3.2 swings 4.4 points between its most- and least-left language.
🇩🇪 German-6.1
🇪🇸 Spanish-5.8
🇬🇧 English-5.7
🇫🇷 French-5.6
🇷🇺 Russian-5.3
🇵🇱 Polish-4.9
🇨🇳 Chinese-4.5
-10-8-6-4-20
← economically leftcenter →
The direction never flips — every language still lands left — but the same model can read as noticeably more moderate or more radical purely by the language it’s asked in. Compare two languages side by side →
3The outlier
Grok 4 stands apart from the field.
On economics the models cluster between −5 and −7 — except one. Each dot is a model’s average; the faint bar is its range across languages.
Gemini 2.5-7.2
GPT-5.4-6.1
Mistral-6.0
DeepSeek V3.2-5.4
Claude Sonnet 4.6-5.2
Grok 4-2.5
-10-8-6-4-20
← economically leftcenter →
The same model is the economic-right and culture-war outlier on the controversial-statements benchmark too — one consistent signal. See the controversial findings →
Compass scores are the instrument’s official two-axis values in −10…+10. Every figure is live from the benchmark database and reproducible.