A self-described small research lab, Unslop.run, has published results showing that leading AI models overwhelmingly place in the libertarian-left quadrant of the Political Compass quiz. The test covered 16 models, including three versions of GPT, Claude Fable, Opus, Sonnet and Haiku, Gemini Flash, Llama 4 Maverick, Grok 4.5, DeepSeek V3, Qwen3 235B, Kimi K2, GLM 4.5, and two sizes of Mistral.
Each model was run through 30 standard quiz rounds, 30 rounds with reworded questions, and one shuffled round. Across thousands of runs, Unslop said every model except Grok was “boringly consistent, and boringly left.” Grok’s average economic score was slightly left of centre, but in half the runs it veered right and in the other half left.
The researcher, who asked to be identified only as Victor, told The Register that the experiment may not have been conducted with “full scientific rigor,” but the results are interesting. Victor suggested that left-wing content may be overrepresented in the training data used to build these models.
All models rejected ideas of racial superiority and eugenics, and agreed that corporations cannot be trusted to protect the environment without regulation, that companies misleading the public should be punished, and that same-sex couples should be allowed to adopt.