AI Gateway lets you rank providers by cost or speed

Who: Posted on the Vercel changelog, authored by Walter Korman. Vercel is a cloud platform company best known for hosting front-end web applications; their AI Gateway product sits between a developer's application and multiple competing AI model providers, routing requests intelligently.

What's new: Vercel's AI Gateway now lets developers explicitly rank which AI provider gets tried first for any given request, choosing between three sorting criteria: cost per million tokens, TTFT, or TPS. Previously, Gateway blended these factors automatically with no user control.

How it works: A developer sets a sort value on what Vercel calls providerOptions.gateway in their request configuration. At the moment a request arrives, Gateway scores all available providers for that model in real time against the chosen metric — cheapest input price, lowest median TTFT, or highest median TPS — and attempts them in that ranked order. Fallback to the next provider only happens if the top-ranked one is unavailable, not on quality grounds. The sort option also composes with Zero Data Retention filtering and with an explicit provider priority list, where hand-picked providers are promoted to the front before the sort criterion is applied to the remainder.

Why it matters: Because rankings are recomputed per request from live data, developers automatically benefit from a provider dropping its prices or improving its infrastructure without touching any code. The routing metadata returned with every response shows which providers were considered, their metric values, and why any were skipped — giving teams an audit trail for cost and performance decisions rather than a black box.