Jev: Vercel’s Cutting-Edge AI Model on the Rise
Jev has reached an extraordinary milestone, becoming the fastest adopted AI model to date according to specific standardized metrics.
Data released by Vercel on Thursday reveals that within just two weeks of its launch, this groundbreaking decision-making model has been adopted by 35% of paid teams—almost double the adoption rate of its nearest competitor, Sonnet 5.5. Vercel’s findings indicate that this marks a new record for the highest adoption rate ever observed for any model during its initial two weeks, highlighting a significant achievement for its developers, TypeSafe AI.
Debuting on September 15 with limited access, Jev is designed as a powerful tool for selecting from a short list of predefined outputs. This approach allows the model to function more efficiently and at a reduced cost compared to alternatives like Sonnet, especially in rapidly categorizing large datasets. The model’s novel methodology has attracted considerable attention since its release, eliciting competitive responses from companies such as OpenAI and Amazon—though substantial evidence to back these reactions has been sparse to date.
The insights come from Vercel’s AI gateway, a unified API service similar to OpenRouter. While it may offer a somewhat narrow perspective—potentially minimizing the impact of major players like OpenAI and Anthropic, which have their own APIs—it confirms that the excitement surrounding Jev has led to tangible applications. Notably, Vercel was an early adopter of Jev, using it to replace GPT-6 Luna for the classification of harmful commands.
Additionally, the report highlights the increasing importance of open-source models, which now account for over 70% of the tokens processed through Vercel’s gateway and nearly 25% of total expenditures.
As open-source models continue to gain traction in the market, the entire sector is experiencing rapid growth. Investment in frontier labs has increased tenfold, although it still represents a smaller portion of overall spending, reinforcing the notion that frontier models and open-source solutions can complement each other rather than vie for dominance.
If you make a purchase through links in our articles, we may earn a small commission. This does not affect our editorial independence.


