Vercel lets teams switch AI models across apps without redeploying

Vercel introduced Virtual Models on AI Gateway on October 28, letting teams change the AI model behind an app or agent without editing its code or deploying it again. Apps call a reusable ID instead of a specific model; Vercel says changes behind that ID reach connected apps within a few minutes.

AI Gateway already let developers switch models by changing one line, while handling provider routing and fallbacks. But that line remained in application code, so a switch usually required a new deployment.

Vercel frames the change around a fast-growing catalog: it says 125 models from more than 10 labs became available on AI Gateway over the preceding 90 days.

A task name instead of a model name

A team creates an ID for a job, such as vmc/coding-agent, and assigns a model and settings to it. Every app or agent performing that job can use the same ID. Teams can create virtual models through their dashboard or the Vercel command-line tool.

Different tasks can have different configurations. Vercel contrasts frequent support-ticket triage, where cost is a priority, with less frequent legal-contract review that needs stronger reasoning.

For coding agents such as Claude Code and Codex, Vercel directs users to connect with this command, then select the virtual model in the agent’s model picker or configuration:

vercel ai-gateway setup

Rules follow the task

Anyone on a team can call a virtual model, while team roles determine who can change it. That lets platform and security teams govern model choices and settings centrally.

The precedence rule also covers data handling. Vercel says requirements for Zero Data Retention or providers that do not train on request data remain enforced even when a request explicitly tries to disable them.

Tags attached to a virtual model replace application-supplied tags, keeping spending attributed to the task. The Virtual Models page displays requests, spending and success rate for each task.

A faster switch still needs testing

Fallbacks, Zero Data Retention settings and tags stay with the virtual model when its underlying model changes. Provider-specific options do not transfer: an Anthropic-specific setting stops applying if the ID is redirected to an OpenAI model. Vercel advises reviewing those options whenever switching providers.

Before changing an ID that serves production traffic, Vercel recommends creating a second candidate ID and running quality tests against the proposed model. If quality drops after the switch, the original ID can be pointed back to the previous model.

Vercel says the feature supports every model type on AI Gateway, including image, video, speech, embedding and realtime models, plus the AI SDK and OpenAI- and Anthropic-compatible APIs.