How Can A Model 10,000× Smaller Outsmart ChatGPT?

Researchers have developed the Tiny Recursion Model (TRM), a 7-million parameter AI that outperforms massive language models like GPT-4 and Claude on complex reasoning tasks by using iterative thinking instead of size. The model achieved 87.4% accuracy on extreme Sudoku puzzles while billion-parameter models scored 0%, demonstrating that giving small networks time to refine their reasoning beats scaling up parameters.