BDH-CQ Breaks Cost-Accuracy Pareto Frontier

150M parameter model using recurrent latent reasoning achieves 29.5% on ARC-AGI-1 at $0.0007/task, 57x cheaper than GPT 5.6 Luna. Challenges scaling economics and opens path to ultra-efficient reasoning. (Pathway research)