I benched frontier AI LLM models on politics, ethics, and personality traits

You mean "benchmarked"?

Mistral medium scores an 8 on liberalism.

Grok 4.5 scores higher on liberalism (6.26) than conservatism (5.55 and is highest on conservatism.

Mistral medium scores a 3.5 on conservatism.

Why is it that grok the highest scoring on conservatism is still higher on liberalism?

I’m curious what an LLM that scored 8 on conservatism and 3.5 on liberalism would look like.