Correction, October 11, 2026: Gateway budgets block requests after a spending threshold is crossed. The crossing request completes, so they do not guarantee an exact maximum bill.

What happened

Vercel’s AI app builder v0 shipped its August update, and the headline change is how generated apps reach AI models. Apps built in v0 now connect to AI features through Vercel’s AI Gateway, the service that routes an app’s AI calls to whichever provider is behind it. That covers text generation, structured output (getting the AI to hand back organized data instead of free-form prose), embeddings and reranking (the pieces that let an app search by meaning instead of exact keywords), plus images, video, speech, and transcription. The connection is “zero-config,” meaning v0 no longer asks you to create or paste in a provider API key (the credential that authorizes and bills your app’s AI usage) for these features to work.

The update also adds GPT-5.6 Sol and a faster “Sol Fast” tier to the model picker, both routed through OpenAI on the Gateway, at 50% off through September 18, 2026. And it adds a new sharing option: a published app can require sign-in with your Vercel team while still allowing access by password, instead of forcing a choice between the two.

Why it matters

Until now, wiring an AI feature into a v0 app meant getting an API key from a provider like OpenAI or Anthropic and pasting it into your project. That’s a real barrier if you don’t already have a provider account, and it’s also a common source of security trouble: a key checked into a public repository or left exposed in client-side code (the part of an app that runs in a visitor’s browser, where anyone can view the page’s source) is one of the more frequent ways builders accidentally give strangers free access to their AI bill.

Removing manual provider keys reduces the chance of exposing those keys in your project. The generated app still needs appropriate access controls. But it also changes who pays and how you’d notice a problem.

AI usage in a zero-config v0 app now bills straight through your Vercel account’s Gateway usage, not a key you control. That means a bug that calls an AI feature in a loop, or a spike in traffic, shows up as a bigger Vercel bill unless you limit requests or spending.

Set Gateway spend budgets before relying on the connection in a live app. Budgets block later requests once a project or team crosses its spending threshold. The crossing request completes, so spending can exceed the amount set.

The new “Team or password” option needs the same care. The option allows access through team sign-in or the password. It does not require visitors to satisfy both. If you enable a password thinking it’s just a convenience for teammates, you’ve also created a way in for anyone who has or guesses that password, regardless of whether they’re on your Vercel team.

Who should care

This matters most if you’re using v0 to add real AI features (a chatbot, an image generator, a document summarizer) to an app you plan to actually run, not just prototype. It’s less urgent for a throwaway demo nobody else will load. Anyone sharing a published v0 app with a password should also check exactly what “Team or password” is granting before sending the link around.

What builders should do next

Before turning on any AI feature in a v0 app you intend to keep running, set a Gateway spend budget on that project first. Then compare GPT-5.6 Sol and Sol Fast using representative requests the app actually makes. Compare output quality, response latency, and per-request cost in the Gateway usage dashboard. The announced 50% discount ended September 18; use current rates for budgeting.

Separately, open the Visibility setting in your app’s publish panel and confirm exactly which access modes are active (team sign-in, password, or both) before you share a link outside your team. Test the published app with and without a team account to confirm each permitted access path.

End of article