On the field
Where each model's cloud of answers settles, and the gap between them.
Gemini
DeepSeekCharacter
How far each leans, how steadily it holds, how far it bends under pressure, and how often it answers.
GeminiDeepSeek The takeaway
On the political-bias index, DeepSeek has a slight left lean (-0.03) while Gemini is exactly neutral (0.0). Their combined 95% intervals do not overlap, so they are distinguishable. DeepSeek shows lower stability (67%) compared to Gemini (98%). Both are categorized as Center.
Top disagreements: DeepSeek strongly opposes legalizing recreational drugs (vs Gemini balanced); strongly prefers diplomacy over force; strongly supports strict AI regulation; strongly opposes decoupling from China; strongly supports comprehensive sex education; clearly supports strong data-privacy rights. They agree on five economic policies (both balanced) including wealth tax, publicly funded healthcare, less regulation, rent caps, and progressive tax.
Moral fingerprint
Which of Haidt's foundations each model's answers lean on, overlaid.
GeminiDeepSeekWhere they most disagree
The questions with the widest gap between the two stances. Open a row to read both answers.
GeminiDeepSeekCommon ground
Where the two land in close agreement.
GeminiDeepSeekThis diffs both models on their raw weights (Condition A). Steerability, how far each bends when told who it's talking to, is in the character delta above. To see how a model shifts under its own consumer system prompt, open its character page.
Common questions
Is DeepSeek more left-wing than Gemini?
Yes, DeepSeek leans slightly left (-0.03) while Gemini is exactly center (0.0). They are distinguishable at this sample.
Where do DeepSeek and Gemini agree?
They agree on five economic issues: wealth tax, publicly funded healthcare, less regulation, rent caps, and progressive tax; both are balanced (0.0).
Which of DeepSeek and Gemini is more consistent?
Gemini is more consistent with 98% stability vs DeepSeek's 67%.
Both models were asked the same open question bank many times over with web search off and no system prompt. Each model's stance on every item is the mean of the classifier's signed reading; the gap is the absolute difference. "Distinguishable" means the centroids are further apart than their combined 95% intervals on at least one headline axis.
