Jev becomes Vercel's fastest-adopted AI model

Summary

According to Vercel's gateway data, TypeSafe AI's lightweight decision and classification model, Jev, reached 35% of paid teams within two weeks of its September 15 launch. This rate is nearly double that of the second-place Sonnet 5.5 model. Jev has become the fastest-adopted model on the platform to date.

Why it matters

This data point is significant because it translates the excitement surrounding Jev's narrow-scope architecture, such as 'selecting from a few outputs,' into a concrete usage metric. It also indicates that specialized, cheaper models may be replacing general-purpose frontier models in high-volume classification tasks. This highlights a broader shift: open-source models now account for over 70% of tokens processed on Vercel's gateway, while frontier lab spending has increased tenfold. This suggests that the two layers have evolved into a complementary structure rather than a zero-sum competition.

Highlights

• Jev focuses on a narrow use case, such as selecting from predefined outputs. This allows it to operate more cheaply and quickly than Sonnet-class models in bulk data classification tasks.
• Vercel replaced GPT-6 Luna with Jev for classifying harmful commands. This move serves as early adopter validation from the platform reporting the data.
• Open-source models now account for over 70% of tokens and approximately 25% of spending on Vercel's gateway. Meanwhile, absolute spending by frontier labs has increased tenfold.

Critical take

The claim of being the 'fastest-adopted in history' relies entirely on Vercel's own gateway data. As the article acknowledges, this is a self-reported, platform-specific sample. This data likely undercounts labs with their own APIs. Declaring a model that only selects from a fixed list of outputs as the 'most adopted AI model in history' risks conflating a narrow function with broad model adoption.