Thinking With Looped Flows Hits 58.8% on ARC-AGI-1, Joining the Looped-Flow Reasoning Wave
alec_helbling · x · 2026-09-17
The FRM authors highlight a converging research direction: looped flow architectures pairing parallel generation with iterative refinement. The new paper Thinking with Looped Flows (arXiv 2609.11801) trains recurrence with local denoising objectives — building temporal association across objectives via progressively decreasing, shared noise — so early updates learn to support later ones despite gradients covering only a few steps. Inference integrates a probability-flow velocity field on a finer temporal grid, and different noise seeds yield multiple valid predictions. Across six reasoning benchmarks it beats prior looped-model SOTA, scoring 58.8% on ARC-AGI-1 and 12.2% on ARC-AGI-2.
More from Research
- Dream-RSI: Google DeepMind paper makes the research process itself self-improving — McDonaghMatthew · 2026-09-17
- Dream-RSI from Google DeepMind turns AI experiment logs into reusable self-improvement testbeds — McDonaghMatthew · 2026-09-17
- MIT's FloatForm robot boats self-assemble into bridges on water, published in Nature — lukas_m_ziegler · 2026-09-17
- DeepMind open-sources a fruit fly that learned to fly, 59-dim action space on a laptop — anthara_ai · 2026-09-17
- XPeng unveils XPACE, a unified embodied world model powering its IRON humanoid — CyberRobooo · 2026-09-17
- Max-pooling attribution maps model outputs back to input words in new sparse-model release — antoine_chaffin · 2026-09-17