A model can spread quickly without becoming a large share of an inference bill. TechCrunch’s new report on Jev’s adoption points to that distinction. The underlying Vercel data offers a more useful story than a single winner’s ranking: adoption breadth, usage and estimated spending measure different aspects of deployment.
The data is particularly interesting because the apparent contradiction is not a contradiction at all. Many teams can try a cheap, narrow service while directing most of their budget elsewhere. Interpreting that pattern requires looking at both the denominator and the kind of work the model performs.
Broad Reach Can Coexist With a Small Bill
Vercel’s October 15 Production Index says Jev reached about 35% of paid gateway teams in its first two weeks, while taking 0.27% of spending during that window. Across September, its reported spend share was 0.15%. The periods differ; those percentages should not be presented as competing estimates of the same interval.
The report also puts open-weight models at 71.5% of gateway tokens and roughly 23% of spend. These are measurements from Vercel’s service, not worldwide market shares. Spending uses published list prices rather than verified customer invoices, and token counts include categories such as cached and reasoning tokens.
An adoption headline therefore cannot establish revenue leadership, quality or business value. Nor does a small share of estimated spend establish that a model is commercially unsuccessful. The figures describe the particular behaviour being counted. Choosing which measure answers the question is part of the analysis, not an inconvenience to remove from the ranking.
The Account Workspace Is Not the Whole Market
Vercel’s separate explanation of its leaderboard data defines a team as an account workspace used by an individual or group. One company can have several teams. A percentage of teams consequently cannot be converted directly into a percentage of companies without additional information.
The same page describes a rolling 90-day view for its geographic and organisation-type breakdowns. At the time reviewed, it attributes 91.4% of token volume to US accounts and 66% to startups. It explicitly cautions that gateway activity does not represent the entire AI market. Those contextual figures also should not be silently relabelled as the September-only composition of the launch cohort.
Location is determined through billing country or, when unavailable, the creator’s signup country. It is not the location of every request or end user. These definitions make the data more interpretable while narrowing the conclusions it can support. A large sample of activity can describe a service precisely without becoming a representative sample of every organisation buying AI.
Decision Models Change What a Request Contains
TypeSafe’s developer introduction explains why Jev warrants a different comparison from a long-form assistant. It takes state and typed questions and returns structured values that software can use directly. A Choice selects from supplied options, a Score evaluates against a rubric and Noul returns a value between zero and one. Some outputs include confidence information.
The documentation recommends decomposing a complex judgement into focused questions and combining the results in application code. These are the developer’s design descriptions, not an independent assessment of accuracy. Their relevance here is that a request count need not describe equivalent work across systems: a short classification and an extended generated response are different units of activity.
None of that establishes which model is best for a particular organisation. Reliability, error consequences and the cost of the complete workflow would require their own evidence. It does explain why reach and spending can move differently without either measurement being wrong.
The new index supplies evidence of actual gateway uptake beyond launch publicity. Its strongest interpretation preserves the distinctions that make the evidence useful: which teams were counted, over what period, for which metric and within which service. Jev’s reach is noteworthy on those terms, without turning a platform-specific adoption result into an industry-wide verdict.
