IAAILEV
LEV: Thinking Fast and Deep with Looped Decision Models
A decision model that answers typed multiple-option questions in one pass but can think deeper by rerunning its layers in a loop instead of generating text. It reads the options after every loop step, so one checkpoint answers at any depth, and a small stopper sets depth per question.
Inside the method
Input + options → [ Looped LM layers ] × r (read options each step) → probabilities
Simplified conceptual schematic. Consult the paper for the complete architecture.
- Recurrence family
- Adaptive depth
- Depth control
- On-demand depth via a learned stopper
- KV / state strategy
- Not applicable (single-pass decision model)
Reading note
A 2.6B LEV rises from 63.2% at depth one to 76.2% at depth four on JevBench. On-demand depth saves 15–24% compute for under one point of accuracy; built on looped pretrained backbones.
Sources checked 2026-10-28. This catalog does not imply independent reproduction.
Related works
Cite this work
@misc{lev2026,
title = {LEV: Thinking Fast and Deep with Looped Decision Models},
author = {Dingjie Song and Zhiling Yan and Yuxuan Zhang and Lichao Sun},
year = {2026},
eprint = {2610.20951},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.20951}
}