Cheap classifiers route decisions across Gemini models

A test with Jev classifier shows atomic decision-making outperforms broad routing or heuristic rules on unseen tasks. Local heuristics degrade significantly under real-world conditions, while Jev maintains 100% accuracy. Setting maxOutputTokens does not reduce token costs for simple tasks, as outputs rarely reach the limit.

Summarised from the original article by an open-source model, then checked automatically: every number and English name must appear in the source. The original article is authoritative.