This New AI Got Popular Fast by Making Quick Choices

Jev, a new AI model from TypeSafe AI, was being used by 35% of paying teams on Vercel within two weeks of its limited release on September 15. That is the fastest two-week pickup Vercel has recorded. TechCrunch That reach was nearly double Sonnet 5.5, the next most popular model in that period.
Growth was fast from the start. Within 24 hours on AI Gateway, Vercel's shared service for using AI models, Jev reached more than twice as many paying teams as any earlier launch. Vercel A September 18 post put early use at about 13% of teams on AI Gateway, about double the early use of the GPT-5.6 family and about six times Fable 5.1. Vercel
Jev is not a chatbot. Think of it as a traffic light rather than a storyteller. It is described as a small tool for picking from a short list of set answers. TechCrunch It is built to return typed decisions, which means clear results that apps can act on right away. Vercel It reads the background it is given and returns picks, scores and odds. Vercel
Vercel uses Jev itself. The company swapped out GPT-6 Luna and now uses Jev to spot harmful commands. TechCrunch After the change, it got answers five to 18 times faster and with better accuracy. TechCrunch Open-source models now make up more than 70% of the data run through Vercel's gateway and nearly 25% of total spending. TechCrunch
The broader context here is about design, not just rivalry. Big creative AI is good at writing and making new things. It is clumsy at fast yes-or-no or pick-one decisions, especially when software needs a tidy answer, not a paragraph to read. Jev fills that gap, background in, decision out, with scores to help send work the right way.
Looking at what this means for teams on AI Gateway, Vercel's own use is the point to weigh. Spotting harmful commands leaves little room for mistakes. Misses bring safety risks. False alarms block normal work. Delays slow everything down. A five to 18 times speedup with better accuracy, if it works the same for others, means more checks can happen right away instead of later or only now and then. That matters for safety checks, tool use and agent loops, where each extra step adds cost and waiting.
In my view, the fast pickup also says something about how developers now use AI. They do not just replace one big model with another. They add smaller, expert models at the edges for sorting, safety checks, ranking and picking. Open-source models already make up most of the data on Vercel's gateway but a smaller share of spending, which fits lots of cheap, repeated helper jobs. Jev fits that pattern. It will be worth watching whether these decision models become basic plumbing, used far more often than the big models they support.


