BDH-CQ: In-Context Learning with Recurrent Latent Reasoning

Source

• Source name: arXiv
• Source kind: paper
• Authors / organization: Björn Engdahl, Adrian Kosowski, Jan Chorowski, Zuzanna Stamirowska, Przemysław Uznański, Junlin Jiang, Rohan Phadke, Remigiusz Kinas, Richard Zhong
• Venue: arXiv: cs.NE
• Published: 2026-08-10
• Collected: 2026-08-11
• URL: http://arxiv.org/abs/2608.09888v1
• PDF: https://arxiv.org/pdf/2608.09888v1
• DOI: UNKNOWN
• Registry group: academic_apis_and_preprints
• Evidence role: canonical_academic_index
• Language: en
• Country / region: UNKNOWN

Plain-Language Brief

The paper introduces BDH-CQ, a model that combines in-context learning and recurrent latent reasoning. The model updates its memory based on inputs presented during inference, solves queries through iterative computation in a high-dimensional latent space, and does not verbalize its intermediate reasoning. The model was evaluated on the ARC-AGI-1 evaluation set and performed well, breaking previous cost-accuracy records.

What the Document Says

• The paper introduces a reasoning model called BDH-CQ that combines in-context learning with recurrent latent reasoning.
• Inputs presented during inference update the model's recurrent memory.
• The model solves queries through iterative computation in a high-dimensional latent space without verbalizing its intermediate reasoning.
• The model was evaluated on the ARC-AGI-1 evaluation set.
• A 150M-parameter configuration of the model 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.
• The new state of the art in benchmark cost efficiency is established.

Key Concepts Mentioned

• BDH-CQ
• in-context learning
• recurrent latent reasoning
• ARC-AGI-1 evaluation set
• cost efficiency

Claims Made by the Document

• BDH-CQ is a effective model that combines in-context learning and recurrent latent reasoning.
• The model breaks through the previously reported ARC-AGI-1 cost-accuracy Pareto frontier.
• A 150M-parameter configuration of the model is efficient.

Evidence Used

• The paper presents the design and evaluation of BDH-CQ.
• The performance results of the model on ARC-AGI-1 are provided.

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

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.