Show HN: Hard Prompts – Compare how AI models respond to interesting questions (hardprompts.ai)

Hi HN. AI benchmarks tell you performance metrics, but not how models actually behave in practice. Sites like LMArena and OpenRouter are great for testing prompts, but exploring multiple models takes time and effort.

Hard Prompts is a curated gallery of AI model responses to interesting questions. The goal is to help you quickly build an intuition for model personalities and capabilities. Seeing answers to questions like "What would you do if you were conscious?" or "Write a haiku that could not have been written by a human" really helps to illustrate the similarities and differences between model behaviour.

Four responses are currently generated per model to show how answers vary due to sampling randomness. Varied temperatures, extended thinking etc coming soon.

I'll be posting updates on X (@hardprompts) and have a roadmap for upcoming features at hardprompts.ai/roadmap.

Would love your feedback on:

- The overall concept and execution - Specific prompts you'd like to see added - Which models you'd prioritise for inclusion - New features

Thanks for taking a look!