TechCrunch reported on 16 October that Vercel gateway data published Thursday shows Jev in use by 35% of paid teams within two weeks of launch, nearly double the reach of the next model, Sonnet 5.5. By Vercel’s own data, that is the fastest adoption a model has seen in its first two weeks on this gateway. The article’s opening limit is “at least by some measures.” It is not a claim about usage across the whole industry.

[1]
A closed binder standing upright, charcoal cover and solid muted-blue spine
A closed report binder standing upright, its cover painted charcoal and its spine a solid muted blue, with pale paper at the top., AI-generated illustration, not a news photograph

TypeSafe AI made Jev, and it was published in limited release on 15 September. TechCrunch describes it as a lightweight decision model that chooses among a few predetermined outputs. Limiting the outputs makes it cheaper and faster than models such as Sonnet, which suits classifying large amounts of data. The buzz had little hard usage data until this gateway report.

[1]

The numbers come from Vercel’s AI gateway, a unified API comparable to OpenRouter. The article says the view is necessarily limited, and is particularly at risk of undercounting labs such as OpenAI and Anthropic that offer their own APIs. Vercel was an early user and replaced GPT-6 Luna with Jev for classifying harmful commands. The 35% therefore includes the gateway operator’s own adoption.

[1]

The same material says open models now account for more than 70% of the tokens processed through the gateway and nearly 25% of total spend. Spending on frontier labs has grown tenfold, while their share of spend has become smaller. The article treats that as a sign that frontier models and open models may be complementary rather than substitutes. Tenfold growth and a smaller share can both be true because the total is growing. All of it is still this gateway’s ledger.

[1]