MineBench Comparisons of a map of the United States

MineBench has released version 4.0.0, now available on iOS, introducing A/B testing for private model checkpoints, and a Community Gallery for custom prompts. Users can showcase and upvote prompts to be added to the official benchmarking set. For a limited time, logged-in users receive unlimited generations with Gemini 3.7 Flash. MineBench is a benchmark that evaluates how well models create 3D Minecraft-like structures from block palettes and prompts, returning JSON coordinates for each block.