I wrote a tool called llmwalk (https://github.com/samwho/llmwalk) that’ll deterministically show you the likelihood the top N answers are for a given open model and prompt. No help on frontier models, but maybe helpful if you want to run a similar analysis more quickly on open models!