Analysis
TypeSafe AI's Jev, a lightweight "decision model" built for high-volume classification tasks, was used by 35% of paid teams on Vercel within two weeks of its production rollout -- nearly double the adoption of the second-most-popular model on Vercel's AI gateway, Sonnet 5.5, and what Vercel calls the fastest model adoption its gateway has ever recorded in a model's first two weeks, according to TechCrunch. Vercel was an early adopter itself, swapping Jev in to replace GPT-6 Luna for classifying harmful commands.
The adoption data lands five weeks after Pulse covered TypeSafe AI's $870 million Series A at a $7.5 billion valuation, itself announced just weeks after a $40 million seed.
“That split suggests Jev's win is specifically in high-volume, low-stakes classification work -- not a sign that cheaper models are displacing frontier labs broadly.”
Jev had a limited release on September 15; by September 18 it was already generating developer buzz, and by month's end both OpenAI and Amazon had shipped their own decision-model clones -- a sign the category moved from novel to contested within about two weeks.
Vercel's broader gateway data adds context: open-source models now account for more than 70% of tokens processed but closer to 25% of total dollar spend, while spending on frontier-lab models has grown roughly tenfold even as its share of the total shrinks. That split suggests Jev's win is specifically in high-volume, low-stakes classification work -- not a sign that cheaper models are displacing frontier labs broadly.
Vercel's gateway view is also a limited lens: it likely undercounts usage at labs like OpenAI and Anthropic that serve customers directly through their own APIs rather than through Vercel's infrastructure, so the 35% figure measures Vercel's specific customer base, not the market overall.