Political Bench is an interactive display of AI-model responses to a political-compass questionnaire. The page presents a dataset labeled December 25, 2025, with 40 models and a chart using economic left-right and social authoritarian-libertarian axes. Its description says each model answered the 62-question test nine times, giving visitors a way to compare the recorded responses within that particular evaluation.
The interface offers a compass view, question view, model-family hulls, cost, speed and methodology tabs, along with a model selector. The chart places the calculated results on two scales running from minus ten to plus ten. These controls make it possible to examine the test from more than one angle rather than rely only on a single point plotted for a model.
The results are a snapshot of responses to a defined questionnaire, not proof that a model has human political beliefs or behaves identically in every conversation. Model versions, prompts, repeated runs and scoring choices can affect an evaluation of this kind. Political Bench is useful for exploring and questioning the displayed dataset, while the test date means it should not be read as a live ranking of every current model. The site credits the benchmark and political-compass resources on which it builds.
