In the AI Gateway Production Index published on October 15, 2026, Vercel reports that Jev reached 35.3% of its external paid gateway teams in the model’s first 14 days. Yet Jev took only 0.27% of estimated gateway spending in that same window. The contrast shows why counting teams, requests and dollars can tell different stories about a model’s use.
What Vercel measured
Vercel’s report ranks September model launches within traffic routed through its AI Gateway. Jev, a TypeSafe AI decision model available through the gateway since September 2026, returns typed choices, scores or Boolean answers for declared questions rather than generating long responses. Vercel says Jev had the highest first-two-week team reach in its gateway history and the most requests of any model in September. Of teams that tried it in week one, 55.6% used it again in week two.
Why the spending number is smaller
A team needs only one call to count toward reach, while spending tracks estimated costs across all paid gateway traffic in the same 14-day window. Vercel describes Jev requests as comparatively short and low-token. That may help explain how a model can appear across many teams and requests yet take a small share of the bill. It does not show whether those calls were accurate, valuable or replacing another model.
What operators should take from it
The denominator is Vercel’s external paid gateway teams, not all AI developers or enterprises. Vercel estimates spending from published list prices, so actual customer bills can differ. Vercel is reporting activity through its own paid product. For a team considering a decision model, this index is evidence of gateway use, not a benchmark: test output quality, uncertain-case routing, latency and cost in the actual workflow before changing production.
