Cloudflare’s Auto Router Rates Each Request for Stakes, from the Most Recent Turns
Cloudflare has released Auto Router in public beta through AI Gateway, picking a model per request from a classifier that scores complexity, ambiguity, stakes and context dependence. Cloudflare’s own benchmark puts it ten points below Claude Opus 5.5 on success rate at a third of the cost, and practitioners have questioned how stakes survive a long agent session. By Steef-Jan Wiggers
Cloudflare has released Auto Router in public beta through AI Gateway, which rates each request for stakes from the most recent turns. The system consists of two stages: first, AI Gateway builds a candidate pool by filtering models based on the request format, credentials, billing configuration, access control, and spend limits.
Then, a multi-head classification model runs on Cloudflare's edge GPUs, assigning probabilities across 14 task categories and rating the request on four dimensions: complexity, ambiguity, stakes, and dependence on earlier context. The scoring matrix combines these factors with benchmark results, using the formula utility = expected quality - adaptive cost penalty.
The Auto Router classifies each request from the newest turns only, with a cap of 8 KiB of text, and scores candidate models on strength by task and expected gain on harder work, with price weighted less as difficulty rises.
Brief written by urgent.news from InfoQ's own syndicated text. Machine-written — may contain errors; check the original before relying on it.
