BDH-CQ: In-Context Learning with Recurrent Latent Reasoning
2026-08-10 • Neural and Evolutionary Computing
Neural and Evolutionary ComputingArtificial IntelligenceMachine Learning
AI summaryⓘ
The authors present BDH-CQ, a new type of reasoning model that learns continuously and updates its memory as it gets new information. Instead of explaining its thought process step-by-step out loud, it works through problems quietly in a complex mental space. They tested it on a tricky question set called ARC-AGI-1 and found it performs better for its cost than previous models. Their study also looked at how well it learns from examples and which ideas are still hard for it.
in-context learningrecurrent latent reasoninglatent spaceARC-AGI-1pass@2cost-accuracy Pareto frontierinference costbenchmark evaluationdemonstrationsmachine reasoning
Authors
Björn Engdahl, Adrian Kosowski, Jan Chorowski, Zuzanna Stamirowska, Przemysław Uznański, Junlin Jiang, Rohan Phadke, Remigiusz Kinas, Richard Zhong
Abstract
We introduce BDH-CQ, a reasoning model that combines in-context learning with recurrent latent reasoning. Inputs presented at inference time continuously update the model's recurrent memory; the model then solves a query through iterative computation in a high-dimensional latent space, without verbalizing its intermediate reasoning. We evaluate the model on the public ARC-AGI-1 evaluation set and use controlled ARC-like interventions to study what it learns from demonstrations, how consistently it applies an inferred transformation, and which concepts remain difficult. A 150M-parameter configuration reaches 29.5% pass@2 at a computed inference cost of \$0.0007 per task. This operating point breaks through the previously reported ARC-AGI-1 cost-accuracy Pareto frontier, establishing a new state of the art in benchmark cost efficiency.