A surprisingly small AI model has outperformed some of the world's largest language models on complex logic puzzles, challenging the notion that bigger is always better in artificial intelligence and opening new paths for efficient reasoning systems.