Pathway Launches BDH-CQ Reasoning Model

Pathway researchers launched BDH-CQ, a 150-million-parameter model combining in-context learning with recurrent latent reasoning. On the ARC-AGI-1 benchmark, it reached 29.5% accuracy at $0.0007 per task, surpassing previous cost-accuracy frontiers. The system separates task learning from solving by updating internal memory without generating intermediate thought tokens.