I believe its none since we are optimizing for latency over reasoning.