Jev is Vercel's fastest adopted model ever

Vercel says Jev was adopted faster than any other model in AI Gateway history, reaching about 13% of teams in its first day. That is 2x the GPT-5.6 family and 6x Fable 5.1 over the same window, according to the company's own gateway data.

Vercel

@vercel

X

Jev was adopted faster than any other model in AI Gateway history.

Guillermo Rauch, who posted the chart, calls both the data and the anecdata shocking. His read is that the spike is less about the product being good and more about demand that had nowhere to go while inference felt expensive.

Rauch's read on why it moved so fast

"People are eager to optimize and put AI in even more places," he wrote, describing the adoption as downstream of the "AI is too expensive/slow" zeitgeist. If he is right, the thing that was holding small teams back was the bill, not the capability. Every idea someone shelved because the token math did not work is now back on the table, which is exactly what a day one adoption curve like this looks like.

What people are actually building with it

Tony Dinh shipped an open source Chrome extension that listens to YouTube audio, detects when a video hits a sponsor segment and skips it in real time. He says it runs at about $0.005 per video, and he shipped it as a bring your own key prototype. That is the same cost order he reported earlier when founders first started pricing classification work with the model.

Dan Shipper says Every has been testing it for about a week, which he notes is unusual because they almost never test new foundation models. His framing is the most useful one for anyone deciding whether to care. The model does not produce words as output, it produces probabilities, so it can sit in the judge seat for jobs that previously needed a Fable-level model. In Every's testing he reports it was 25x faster and 600x lower priced than that alternative, and he points to a vibe check by @hammer_mt published on Every.

All of these are the vendors' and users' own numbers, not independent benchmarks. Vercel is reporting on traffic through its own gateway, which measures how many teams tried a new model, not how many kept it in production. Shipper's 25x and 600x come from one company's internal testing on their own workloads.

The business read is that a classifier priced like this changes what is worth automating rather than what is possible. Sponsor skipping was always technically doable. At half a cent a video, one person can ship it as a weekend prototype and give it away.

Guillermo Rauch

@rauchg

X

The data and the anecdata on Jev's adoption are shocking.

Dan Shipper

@danshipper

X

it doesn't produce words as output, it produces probabilities