Flow Reasoning Models hit 99.5% on Sudoku-Extreme with 44× fewer inference FLOPs
mark_k · x · 2026-09-05
Flow Reasoning Models (FRMs) take a new approach to AI reasoning: instead of generating a solution sequentially, the model repeatedly refines the entire solution, reconsidering and correcting its own decisions until convergence. Results: 99.5% on Sudoku-Extreme, 100% on Zebra, 99.9% on Maze-Unique — matching the 98.7% solve rate of the next-best Sudoku method with 44× fewer inference FLOPs. Iterative self-refinement may scale reasoning without brute-forcing inference compute.
Related event: Flow Reasoning Models Rewrite Full Solutions, Cutting Compute 44x(2 posts)→
More from Research
- Rumor: Anthropic's model solved a Millennium Prize Problem, Terence Tao reacts — IgorCarron · 2026-09-05
- AREX-Skill boosts MLE-bench medal rate from 31% to 73% with distilled agent skills — TheTuringPost · 2026-09-05
- Shanghai AI Lab's Harness-of-Harness runs 70+ iterations to build an FPS game autonomously — aigclink · 2026-09-05
- Hugging Face surveys 16 open-source RL libraries: async disaggregation is the consensus — Thom_Wolf · 2026-09-05
- CoRL 2026 paper GRA: synthetic robot videos should supervise geometry, not control — MikeShou1 · 2026-09-05
- RL Hill-Climbing Creates Model Spikes, So Judge AI With Multiple Models — dejavucoder · 2026-09-05