New Paper: 150M Model Achieves 29.5% on ARC-AGI at $0.0007 per Task

burny_tech · x · 2026-08-15

The paper 'BDH-CQ: In-Context Learning with Recurrent Latent Reasoning' proposes a method that updates a recurrent memory with examples and reasons in a continuous latent state without verbalizing intermediate steps. A 150M model reaches 29.5% on ARC-AGI-1 at a cost of $0.0007 per task, setting a new cost-efficiency frontier.

Original post →

More from Research

Research channel →