LEV: Looped Decision Models Hit 76.2% on JevBench by Rerunning Layers, Not Generating Tokens
Dingjie Song · hf · 2026-10-19
Researchers from Lehigh and UBC present LEV, a decision model that thinks deeper by rerunning its layers in a loop instead of generating tokens: one forward pass yields per-option probabilities, and layers iterate with options read after each step, jointly trained so one checkpoint answers at any depth. On three looped backbones accuracy rises with depth, mostly on hard questions: a 2.6B LEV goes from 63.2% at depth one to 76.2% at depth four on JevBench. A threshold stopper fitted on unlabeled requests saves 15–24% computation for under one accuracy point; on a locked test set 1.4B/3B models beat dense models and cascades of the same latency (<50ms) by 5.4–8.8 points.
More from Research
- 1.9 MB extracted (not trained) LoRA fixes vertical color stripes in MOSS-Image SFT — Nynxz · 2026-10-19
- Three ways to execute a AND b AND c: DataFusion filter PR explained — Abhishekcur · 2026-10-19
- Why safety reports can't prove safety: the adversarial logic flaw — davidmanheim · 2026-10-19
- ACT3D trains VLMs to actively explore 3D worlds with RL, hitting 53.2% accuracy — UCSanDiego · 2026-10-19
- OPMD: ShanghaiTech distills teacher intermediate computations beyond output supervision — ShanghaiTech · 2026-10-19
- LexFrame: interpreting LM residual streams in the model's own semantics — Jian Gu · 2026-10-19