Cloudflare Auto Router weighs model choice against cache costs

Cloudflare’s Auto Router classifies each request, estimates task difficulty and balances model quality against cost. It also accounts for the expense of rewriting context when switching models, rather than choosing solely by token price. The public beta exposes this through AI Gateway, but its own evaluation shows that lower cost can come with a lower success rate.

Cloudflare · Sep 30, 2026

Why it matters

Agent teams can evaluate routing against cost per successful task, including context reuse.

More context

Try cloudflare/auto through AI Gateway on a representative task suite. Routing is free during beta; this does not mean underlying inference is free. Compare outcomes and retries before using it for important workflows.

Keep in mind

Cloudflare’s internal evaluation is not an independent guarantee. Full Responses API support, WebSockets and zero-data-retention filtering are listed as future work. 1min has not tested routing quality.

Original read