TypeSafe's Jev became the fastest-adopted model in Vercel Gateway history

TypeSafe is a startup founded by Diogo Almeida, a former OpenAI contributor who worked on ChatGPT. On September 15, 2026, it launched a model called Jev, which the company calls a System One model: fast, narrow, and built for one job rather than open-ended conversation. Jev takes a program's current state plus a typed question, a yes or no, a multiple choice, or a numeric score, and returns a calibrated probability in a single parallel pass, without generating a word of prose.

TypeSafe claims Jev is 20 to 200 times faster than routing the same structured question through a full language model, and 40 to 400 times cheaper. On TypeSafe's own benchmark, Jev scores 88.3 percent accuracy. Those are the company's numbers, not an independent lab's, and they deserve the same scrutiny any vendor benchmark does. TypeSafe frames the tradeoff plainly: give up open-ended generation, and get a narrow answer back fast enough and cheap enough to call on every step of a pipeline instead of a handful of steps, the difference between checking one gate before a response and checking every gate along the way.

Adoption moved fast enough that Vercel noticed. Vercel reported that within 24 hours of launch, Jev reached roughly 13 percent of paid teams on its AI Gateway, the routing layer many production apps use to call different models without rewriting code for each one. Vercel's own blog and changelog both described it as the fastest-adopted model in the Gateway's history. Jev is also live on Cloudflare Workers AI, listed directly in Cloudflare's developer docs, putting it on a second major deployment platform on day one.

What Jev is not is a general-purpose chat model, and TypeSafe is not positioning it as one. The company's argument is narrower: most of what gets called agent decision-making in production is fast structured classification, a yes or no, a category, a score, not open-ended writing, so a model built only for that slice can beat a full LLM on both speed and cost for that slice alone. Whether that holds outside TypeSafe's own benchmark, across real production agent workloads rather than a curated test set, is not yet independently confirmed.

Why a build studio cares

GattyWorks runs builds on both Vercel and Cloudflare, and AI workflows and custom agents are one of our four service lines, so a new model option on both platforms at once is not background noise for us. The number itself is not the point. It is the reminder that a lot of what a client calls an agent decision, approve or reject, which category, how confident, is a structured classification question wearing an agent costume, not a writing task. The next time a pipeline step is really just approve, reject, or score rather than open-ended writing, Jev is now a named, benchmarked option to test against, on infrastructure we already deploy on, not a hypothetical model from a pitch deck.

Next step: read Vercel's blog and changelog for the Gateway adoption numbers, check Jev's listing on Cloudflare Workers AI, and write to us at hello@gattyworks.com.