Jev: ~50× faster, over 100× cheaper. But is it any good?
A new kind of AI model went viral this month. We put it on two of our own tasks against Claude, using real formalization errors from Prover projects.

Prover Labs
Notes from Prover's AI team as the work happens: what we are building, what we measured, and what we learned along the way. Each report is dated and written by the people who did the work.
4 reports · latest 23 September 2026
A new kind of AI model went viral this month. We put it on two of our own tasks against Claude, using real formalization errors from Prover projects.
On a customer project, an AI assistant now works beside the engineer's Prover iLock work. It reads the model and the verification report, and asks Prover iLock to run. It cannot change a single line of the model.
Before building Ask Prover, the question-answering assistant on Prover Labs, we had to choose how it finds things. We measured both ways in one afternoon, on the same documents and the same ten questions, and the measurement decided it.
How Prover Labs verifies its own changes: a strict mechanical gate, then an independent reviewer that drives the app in a real browser, with a verdict that expires when the diff changes.