Jev is Vercel’s fastest-adopted AI model ever

Jev, a lightweight decision model from TypeSafe AI, has become the fastest-adopted AI model in Vercel's data, reaching 35% of paid teams within two weeks of its launch. Vercel published the figures on Thursday. The model went out in limited release on September 15. In the same two-week window, Jev reached nearly double the share of the second most popular model, Sonnet 5.5, according to Vercel.
The measurement comes from Vercel's AI gateway, a unified API service comparable to OpenRouter. That vantage point has a built-in limitation: it is likely to undercount major labs such as OpenAI and Anthropic, which run their own APIs and therefore capture usage the gateway never sees. Even with that caveat, the numbers indicate that the attention around Jev converted into real production usage rather than staying a topic of discussion. Vercel itself was an early adopter, swapping out GPT-6 Luna to handle classification of harmful commands.
The design choice behind Jev is the part developers will care about most. Rather than generating open-ended text, the model is built to pick between a few predetermined outputs. Constraining the answer space that way lets it run cheaper and faster than broader alternatives such as Sonnet, which makes it a natural fit for sorting or labeling large volumes of data quickly. In practice, this is the classic trade-off of a classifier-style model: you give up flexibility in what the model can return in exchange for lower cost and lower latency on a narrow task. That makes Jev a poor substitute for a general assistant and a plausible one for pipelines that need a label attached to each record.
The buzz around the approach has already drawn responses. OpenAI and Amazon, among others, have produced competing offerings in the weeks since Jev's release, according to the report. Until now, there had been little hard data behind the enthusiasm; the Vercel figures are one of the first concrete signals of adoption.
A second thread in the same dataset concerns open source models, which now handle more than 70% of the tokens processed through Vercel's gateway while accounting for nearly 25% of total spend. The gap between those two figures is the interesting part: open models carry the majority of traffic by volume but a much smaller slice of the money. This suggests a pattern where open weights are used for high-volume, cost-sensitive work while paid frontier models are reserved for the requests that justify a higher price per call.
Spending on frontier labs has grown tenfold over the period covered, even though those labs now command a smaller share of overall gateway spend. The report reads that as evidence that frontier models and open source alternatives may end up complementary rather than directly competitive. Read another way, inference demand is expanding fast enough that both sides can grow in absolute terms while the mix shifts. A likely trade-off for teams is added architectural complexity: routing different tasks to different models, tracking spend separately, and accepting that a model chosen today may be the wrong economic fit in a few months.
Why it matters: teams building classification or routing features now have a documented signal that narrow, constrained-output models can win real adoption fast, which makes them worth benchmarking against general-purpose models on cost and latency rather than capability alone. It also means product and platform teams should expect more of these limited-output offerings from the major labs, and plan for a gateway setup that can swap models without rewriting application code. For anyone weighing open weights against frontier APIs, the token-versus-spend split is a reminder that the two are often used for different jobs in the same stack. Vercel's view is necessarily partial, so treat the 35% figure as a strong directional indicator rather than a market-wide share.
Based on reporting from the original publisher. Visit the source for full context and later updates.
Publisher excerpt
The data comes from Vercel's AI gateway, a unified API service similar to OpenRouter.