Jev reached 35% of paid teams using Vercel’s AI gateway within two weeks of its launch, according to data published by Vercel. The decision model’s reported uptake was nearly double that of the second most popular model, Sonnet 5.5.
A model built for bounded choices
Jev was published in limited release on September 15. It is designed to choose among a few predetermined outputs, a narrower task than generating an open-ended answer.
That constraint is central to the model’s pitch. By limiting the available outputs, Jev can run more cheaply and quickly than alternatives like Sonnet. The design makes it suited to sorting or classifying large quantities of data, where a system needs to select from defined categories.
The approach drew attention after release and prompted competing offerings from OpenAI and Amazon, among others. Vercel’s usage figures provide evidence that interest extended beyond discussion: paid teams were using the model through its gateway.
What Vercel’s figures capture
The figures come from Vercel’s AI gateway, a unified API service similar to OpenRouter. They offer a view of model use through that service, rather than a complete picture of the AI market.
The data may undercount major labs such as OpenAI and Anthropic, which offer their own APIs. As a result, the reported adoption rate should be read as a measure of usage among Vercel gateway customers, not as a universal ranking of every AI model.
Vercel itself was an early Jev adopter. The company used the model to replace GPT-6 Luna for classifying harmful commands, illustrating one kind of narrowly defined task for which a decision model can be applied.
Open-source models gain share as spending grows
Vercel’s report also describes a larger role for open-source AI models. They account for more than 70% of the tokens processed through the gateway and nearly 25% of its total spend.
Those shares describe different measures. Token volume reflects how much text or other model input is processed, while spend tracks the money paid for that usage. Together, the figures show that open-source models can handle a large portion of gateway activity while representing a smaller portion of its cost.
At the same time, spending on frontier labs has grown ten-fold, even though it makes up a smaller share of overall spend. A falling share does not necessarily mean lower spending: the total market can expand while a category’s proportion declines.
Different models may serve different needs
The combination of rising open-source use and increased frontier-lab spending supports the possibility that the two groups of models will be complementary rather than competitive. Teams may choose among them based on the task, cost, speed, and output flexibility they need.
Jev’s early adoption offers one example of a specialized model finding a place in that mix. Its measured reach is notable, but the figures come from a single gateway and cover the first two weeks after launch. They show a fast start within Vercel’s service, while leaving broader market adoption uncertain.