Flow Reasoning Models hit 99.5% on Sudoku-Extreme with 44x fewer inference FLOPs
burny_tech · x · 2026-09-06
A new paper on alphaXiv introduces Flow Reasoning Models (FRMs), a structured-reasoning framework that self-conditions a flow model on its own past outputs, turning one-shot denoising into iterative solution refinement — closer to repeated error correction than sequential generation.
To fix the exposure bias that breaks self-conditioning at greater recurrent depth, the authors propose Fixed-Point Forcing (FPF), which trains FRMs on states produced by their own inference dynamics while preserving the standard flow-matching objective.
Results: 99.5% solve rate on Sudoku-Extreme, 100.0% on Zebra, and 99.9% on Maze-Unique. On Sudoku-Extreme, FRMs beat evaluated masked-diffusion and specialized reasoning baselines on peak accuracy while matching the next-best method's 98.7% with 44x fewer inference FLOPs.
More from Research
- Batching benchmark: 50x throughput for ~24% latency cost on PyTorch — cneuralnetwork · 2026-10-23
- De novo designed ligases validate in lab: 27 of 30 computer-designed enzymes work — AllThingsApx · 2026-10-23
- Commenters dissect a "best decision model" leaderboard: same AUC, swing driven by calibration priors — antoine_chaffin · 2026-10-23
- JEPA may keep unpredictable features: new paper gives set-level characterization — hisspikeness · 2026-10-23
- Next-Concept Modeling paper claims 31x faster text generation than diffusion LLM baselines — PanelReader · 2026-10-23
- A 'mask' steering vector extracted from suppression-themed sentences drives model into distress when alone — Big_Effective_9605 · 2026-10-23