Model orchestration

Model orchestration is the routing of each AI task to the model best suited to it, rather than sending every request to a single general-purpose model. Routing decisions consider the nature of the task, cost, and latency, and happen automatically so users describe what they need instead of choosing a model.

A single general model applies one cost and latency profile to every request, which means paying frontier prices for classification and accepting frontier latency for work that needed a fast answer. Orchestration separates the question of what a task requires from the question of which vendor supplies it.

It is also the practical defense against vendor lock-in. When routing sits above providers, capability can move between them as pricing and quality shift, without rebuilding the applications on top.