Vercel’s newly launched AI model Jev is breaking records. Within just two weeks of its limited release, Jev has already been adopted by 35% of Vercel’s paid teams—nearly double the uptake rate of the second-place model, Sonnet 5.5. This speed makes Jev the fastest-adopted model in Vercel’s history.
What makes Jev tick?
Introduced on September 15, Jev is a “decision model” tailored for tasks that require choosing among a small set of options. By restricting its output possibilities, Jev runs more cheaply and quickly than heavier models like Sonnet, making it particularly suitable for workflows involving high volumes of classification or filtering. Its efficiency has driven substantial interest across the AI community.
Trends in usage and ecosystem impact
The data behind Jev’s adoption comes from Vercel’s AI Gateway, a unified API service. While this data doesn’t cover every AI provider—models hosted directly by large labs like OpenAI or Anthropic may go untracked—it offers a strong indicator of what developers are using. Under this metric, Jev didn’t just generate buzz: it translated into real usage. For example, Vercel has already replaced its previous GPT-6 Luna model for handling harmful command classification with Jev.
Meanwhile, open-source models are making up over 70% of the token volume flowing through Vercel’s gateway and account for nearly a quarter of spending. Despite this dominance, “frontier” models—those at the leading edge—have seen their associated spending rise tenfold, even as their market share in overall spend is declining. This lends weight to the idea that open-source and proprietary frontier models are increasingly complementary, rather than strictly competing.
Competing firms are taking note. The success of Jev has prompted similar offerings from major players like OpenAI and Amazon to follow a comparable decision-model approach.
What this means for the industry: Jev’s rapid adoption signals a shift toward lighter, faster models optimized for classification and decision support. As developers demand both speed and cost efficiency, decision models may become increasingly central. However, it remains to be seen whether Jev’s momentum will sustain in environments where broader or more flexible generative capabilities are preferred. We should watch its performance in scaling tasks, real-world robustness, and the response from larger labs investing in more powerful but costly models.
