A preprint compares neural networks and reduced-order models for reacting jet flows, finding task-dependent accuracy and predictions at new spacings.