TRM

7M-Param Model Beats 1000x Larger Rivals: AI Might Not Need Endless Compute

7M-param model beat 1000x larger rivals on ARC Prize via recursive reasoning. It defies 'bigger is better,' letting AI scale without stacking compute.