Pathway's BDH-CQ: Post-Transformer Breaks ARC-AGI-1 Cost-Efficiency Frontier

Pathway's BDH-CQ, a 150M-parameter post-transformer model, achieves 29.5% on ARC-AGI-1 at 11x lower cost than GPT 5.6 Luna via latent recurrent reasoning. This challenges the assumption that chain-of-thought is necessary for reasoning and opens a new cost-efficiency frontier. Validates that architectural innovation, not just scale, drives AGI progress. Watch for scaling results and extension to harder benchmarks.