Jev reached 35% of Vercel’s paid teams in two weeks

What Jev is built to do

Jev is designed for a narrow task: selecting from several predetermined answers. That makes it suited to classifying large volumes of data, rather than handling open-ended requests.

The model drew attention after launch, and OpenAI and Amazon introduced similar approaches. But until Vercel published usage figures, there was little concrete evidence that attention to Jev was translating into broad use.

What Vercel’s data can show

The figures come from Vercel’s AI Gateway, a unified API service similar to OpenRouter. They do not represent the whole market: the gateway may undercount large labs such as OpenAI and Anthropic, which offer their own APIs. Still, the data provides evidence that teams have started using Jev.

Vercel was among Jev’s early adopters, using it to classify malicious commands in place of GPT-6 Luna. Within two weeks of the limited release, 35% of Vercel’s paid teams had connected the model.

The same report points to a broader shift in the gateway’s traffic and spending:

Those figures fit a market where open models and leading labs can serve different needs rather than simply displacing one another. My read is that Jev’s early uptake matters less as a verdict on model quality than as evidence that a focused, cheaper tool can spread quickly inside an engineering organization.

What the data does not tell us is whether those teams keep using Jev, or how much work it handles after the initial connection. Vercel’s figures establish adoption, not sustained use. That gap matters: if open models are becoming the default for narrow tasks, leading labs may keep growing in absolute spending while losing share of the everyday work.