“The low-latency networking of Google Kubernetes Engine (GKE) saves Baseten a couple dozen milliseconds per turn between models in multi-model agentic systems.”