Cloudflare's Auto Router picks an AI model for each request in public beta
Teams get a free way to route requests across models, but they must weigh a roughly ten-point drop in reliability against lower spend.
Cloudflare has put its Auto Router into public beta via AI Gateway: setting the model to cloudflare/auto makes a classifier choose a model for each request. The router first narrows a candidate pool by format support, credentials, billing, access rules and spend limits, then uses a multi-head classifier running on Workers AI to score each request across 14 task categories and on complexity, ambiguity, stakes and reliance on earlier context. Cloudflare's benchmark, covering 97 tasks, reports a model mix that costs roughly a third of the baseline spend while losing about ten points of reliability, which leaves teams to weigh that tradeoff themselves. Practitioners flagged that stakes are judged from recent turns only, so a risky action set up early in a session could be routed as if low-risk, and that the default pool and scoring weights may drift as models change. The beta is free with no post-beta price given, and WebSockets are not supported.
Why it matters
The router scores each request on complexity, ambiguity, stakes and reliance on earlier context, then chooses a model, so the choice of model moves from the developer to the router. Because stakes are judged only from the newest turns, a risky action set up earlier in a session may be treated as low-risk, which matters most for anyone relying on it for sensitive work. The router is free during the beta and no price has been set for afterward, so the long-term cost of depending on it is still unknown.
Matters or hype?
Does this story matter, or is it hype? Vote before you see what everyone else said.