3 AI Models Compete to Make The Best Game (Super Mario Edition)

Can AI models recreate Super Mario level 1? Watch ChatGPT, Grok and Claude compete to build a classic game. I tasked these language models with building the first level of Super Mario from scratch using a single prompt. By forcing these agents to work within the Pixel Lab API, I can observe how they handle art, game mechanics, and the overall feel of a platformer. It is a practical look at how these systems manage complex creative engineering tasks when given specific constraints. The results highlight the specific strengths and weaknesses of each model when applied to game development. While Grok struggles with the implementation, scoring a 3.7/10, Claude excels with a 9.7/10, demonstrating superior logic capabilities and code generation. This AI gaming experiment provides a clear look at which models can handle architectural decisions versus those that fail to maintain project cohesion. Subscribe for weekly AI model comparisons and leave a comment telling me which game you want to see them play next.