According to data released by Vercel, the lightweight decision-making model Jev has been adopted by 35% of paid teams within two weeks of its launch, achieving a market penetration nearly double that of the second most popular model, Sonnet 5.5, setting a new record for the fastest adoption rate in the first two weeks post-launch among models tracked by Vercel. Jev was initially launched on September 15th with limited availability, primarily designed for selecting from a small set of predefined outputs—reducing operational costs and enhancing processing speed by constraining output ranges, making it well-suited for large-scale data classification.

This data originates from the Vercel AI Gateway, a unified API service similar to OpenRouter, which does not encompass major self-hosted APIs; thus, it does not represent the entire market but indicates that Jev’s rising popularity has translated into tangible usage. Vercel previously used Jev to replace GPT-6 Luna for categorizing harmful instructions. Concurrently, open-source models accounted for over 70% of tokens processed via the Vercel Gateway and nearly 25% of total spending, while expenditures at cutting-edge model labs have increased approximately tenfold.