What multi-model routing actually saved us (with real numbers)

Not every step in an agent workflow needs your most expensive model. Classifying intent, extracting a field, or formatting output rarely benefits from frontier-level reasoning.

By routing lightweight steps to a smaller, faster model and reserving the expensive model for genuinely hard reasoning steps, teams on Arclight have seen inference costs drop meaningfully without any change to output quality.

The routing rules are simple to set up and don't require rewriting your agent logic — you're changing which model handles a step, not what the step does.