Why it matters

LLM routing shapes cost + quality. Understanding shapes strategy.

Advertisement

The architecture

Classifier / heuristic picks model.

Cheap for easy; large for hard.

Fallback on failure.

LLM routing flowRequestuser inputClassifier / heuristicsmall / large modelRoutewith fallbackRouter models (Martian, RouterEval) predict best model per query
LLM router.
Advertisement

How it works end to end

Cascade: small first, fallback large.

Router models (Martian, Not Diamond).

Task classification.

Cost + quality metrics.