I benchmarked Liquid AI's new open decision models, d1-3B and d1-omni-600M, as local prompt-injection gates for my Hermes agent, along with Decision-2.0-Lux-9B in MLX. d1-3B is quick on my M2 Max (0.3...