BDH-CQ reaches ARC-AGI at $0.0007 a task

BDH-CQ, a 150-million-parameter recurrent latent-reasoning model, reports 29.5% pass@2 on ARC-AGI-1 at a computed inference cost of $0.0007 per task. The new arXiv preprint calls that a new cost-accuracy Pareto point, but it is a single evaluation and not evidence of broad production capability; benchmark details and independent reruns still matter, as the archive’s reproducibility gate for AI mathematics argues. Model buyers should add cost per solved task to capability comparisons instead of treating parameter count as the budget proxy.