Energy-Based Reasoners cut inference compute 345x, hit 99.9% on Sudoku-Extreme

du_yilun · x · 2026-11-18

willchen1 introduces Energy-Based Reasoners (EBRs), arguing looped reasoning models would work better as energy-based models (EBMs). Key claims: EBRs use 345x less inference compute and 2.6x fewer training FLOPs than the best baselines, are 7x-43x more sample efficient than looped baselines at inference, reach 99.9% on Sudoku-Extreme, and near 100% accuracy on multi-solution benchmarks. Details in the linked thread.

Related event: Energy-Based Reasoners Cut Inference Compute 345x, Solve 99.9% of Hard Sudoku(3 posts)→

Original post →

More from Research

Research channel →