Classical Energies Learn to Reason 1,000x Faster than Looped Models
arXiv cs.AI · 13 Nov 2026, 04:56 AEST
Classical Energies have demonstrated the ability to reason 1,000x faster than traditional looped models. This advancement highlights the potential of energy-based methods for efficient problem-solving in reasoning tasks.
Key facts
- Looped models have emerged as a prominent paradigm for reasoning, with successes on simple reasoning puzzles such as Sudoku-Extreme and Maze-Unique.
- Descent on an energy not only mitigates these issues but also offers advantages: built-in early stopping, compute that adapts to problem difficulty, fast training without differentiating through a solver, and interpretable weights that learn the constraints of the puzzle.
- Here we demonstrate this with a learned version of Hopfield & Tank's energy from the 80s, which we term CRAM, solves 99.996% of Sudoku-Extreme and 100% of Maze-Unique test puzzles after 42 ms and 132 ms of training on one GPU, at least 1,000x faster than looped models on these tasks.
- CRAM also matches baseline looped models at their highest accuracy on Sudoku-Extreme using at least 282x fewer inference FLOPs.
Read aloud
Classical Energies have demonstrated the ability to reason one000x faster than traditional looped models. This advancement highlights the potential of energy-based methods for efficient problem-solving in reasoning tasks. Looped models have emerged as a prominent paradigm for reasoning, with successes on simple reasoning puzzles such as Sudoku-Extreme and Maze-Unique. Descent on an energy not only mitigates these issues but also offers advantages: built-in early stopping, compute that adapts to problem difficulty, fast training without differentiating through a solver, and interpretable weights that learn the constraints of the puzzle. Here we demonstrate this with a learned version of Hopfield & Tank's energy from the 80s, which we term CRAM, solves ninety nine point nine nine six percent of Sudoku-Extreme and one hundred percent of Maze-Unique test puzzles after forty two ms and one hundred thirty two ms of training on one G P U, at least one000x faster than looped models on these tasks. CRAM also matches baseline looped models at their highest accuracy on Sudoku-Extreme using at least 282x fewer inference FLOPs.
Sources in this cluster
- Classical Energies Learn to Reason 1,000x Faster than Looped Models
arXiv cs.AI · 13 Nov 2026, 04:56 AEST