Zachi ends pg-jev over query speed above 50k rows
The Postgres extension that put Jev calls inside queries is done, and the author says the ceiling is scale, not bugs.
Zachi called the newest pg-jev release its last, the Postgres extension that let you call Jev from inside a SQL query. His reason is scale. Under 50k rows it is fine, he says, and above that query speed becomes the bottleneck.
its safe to say that pg-jev wont work as exptected
The number behind the call is roughly 3b tokens spent on benchmarks and testing. That is his own figure, from his own runs, not a published eval, and he does not break out which queries or which hardware. What he does give is the shape of the failure, which is the useful part. It is not a bug list. It is a line on a graph that stops going where he wanted it to go.
What comes next
Zachi says he is building a replacement right now, again on Jev, and that it will be faster, smarter and usable in the database of your choice, naming Supabase, Neon and Drizzle. No release date, no benchmark, no preview numbers. Treat it as a stated intention from the person who just retired the previous attempt.
For people building on it
This is the second time in a week the same ceiling has shown up from a different direction. Shreya Shankar earlier argued that calling a decision model once per row is a bad idea for batch work because it gets no benefit from query planning, and put the speed-of-light for one AI filter over 5k movie reviews at about 6.6 seconds with Qwen3-4B on one H100, a bar she says no system today comes close to, including her own Quail. Zachi hit the practical version of that argument with a row count attached.
So if you were planning to drop a Jev call into a SELECT and let it ride, size the table first. Under 50k rows one builder found it workable. Above that, on his testing, the per-row pattern is the thing that breaks, and swapping models will not move it.
Know what Jev is good at before you bet on it
One email a day. Under three minutes. What builders found out about Jev yesterday.
