Vercel AI Gateway Adds One Fast-Mode Switch

Vercel added a single fast-mode switch to its AI Gateway beta, letting one speed setting apply across every model the gateway routes to, per the July 30 changelog entry.

Anthropic, OpenAI, and xAI each expose latency tiers through their own flags and defaults. Teams routing across providers wrote a branch per provider just to ask for the fast version of a model. Or they skipped the tier, because wiring it up for every provider wasn't worth it.

The gateway collapses that into one request shape. Set speed to fast, and the gateway routes the fast tier to whichever provider handles the call. It translates one flag instead of three separate ones. No per-provider branch, no skipping the tier to save the trouble.

It's a small fix, but it's the kind that matters once you've built a routing layer instead of hardcoding a single model. Anyone who has shipped a fallback chain across two or three providers has hit this wall. The models are interchangeable in principle. The latency knobs are not.

The bet here: this pushes teams further into treating the gateway, not the model provider, as the integration point. If Vercel keeps normalizing provider-specific quirks like this one, switching providers gets cheaper. Switching gateways gets more expensive, the opposite of what multi-provider routing was supposed to buy you.