VeilPlays

Compare AI models through game-based benchmarks

Artificial Intelligence

Tech

Games

OpenAI Day

Visit WebsiteSee on Product Hunt

Hi Product Hunt! I built VeilPlays because comparing AI models through benchmark scores alone can feel abstract and disconnected from what users actually experience. VeilPlays turns AI model evaluation into an interactive blind test. Multiple AI models are given the same prompt and must create a complete playable game in a constrained, single-shot generation process—without follow-up prompts, manual revisions, or additional development requests. Their identities remain hidden while you play. Not every model successfully completed the game under these constraints, and some outputs were only partially functional. I kept those outcomes as part of the experiment, because the ability to deliver a complete, playable result in a single attempt is itself an important measure of model performance. After playing, you can compare and rate the results based on gameplay, visual design, creativity, and code quality—without being influenced by model names or brand reputation. One challenge I’m exploring is how to turn casual gameplay into meaningful human evaluation. Many people enjoy trying the games, but the ratings are what make the blind comparison genuinely useful. So when you try VeilPlays, I’d really appreciate it if you could complete at least one evaluation after playing. It only takes a moment, and your ratings directly contribute to the comparison results. I’d also love your feedback on the blind evaluation experience, the rating criteria, and which AI models or game categories you’d like to see next. Thanks for checking it out!