Definition
Plain language
A compact open-weight language model that loops its layers to get more reasoning out of a small network.
As stated in the literature
A 1.4B-parameter looped attention model used as a backbone in depth-recurrence and consolidation studies, retrofittable with SSM layers for hybrid experiments.
Why it matters: Looping layers offers more reasoning capacity per parameter, which is attractive when you can't afford a larger model.
For example, Ouro reuses the same transformer layers several times in a loop, letting a 1.4B model reason as if it were deeper.
Heard on the show
“And there's Ouro, which is a 1.”Episode 085 — Why Long-Context Models Might Need Compute, Not Capacity, Before Eviction