MotherDuck ships prompt_jev; TypeSafe pauses signups
MotherDuck says 100k rows classified in 40 seconds for $0.50, against 32 minutes and $37 for the LLM it replaced. Hours later TypeSafe suspended new Jev signups, so the people who can act on that number are the ones already in.
Built with it
Hamilton Ulmer shipped prompt_jev() as a SQL function in MotherDuck
Hamilton Ulmer launched prompt_jev(), which runs Jev classification inside MotherDuck SQL, including filtering on meaning in a WHERE clause. MotherDuck's own numbers: 100k rows in 40 seconds for $0.50, against 32 minutes and $37 for the LLM, at what they call frontier-LLM accuracy.
100k rows: 40s, $0.50, frontier-LLM accuracy. The LLM took 32 min and $37.
Someone measured it
TypeSafe paused new Jev signups, existing accounts keep working
TypeSafe AI temporarily suspended signups for Jev, saying demand had swelled enough that it needed to protect quality of service for existing signups. Those accounts continue to function, and the company says it is working to reopen access. Diogo Almeida framed it as the trade-off that lets the team sleep.
our service is absolutely overflowing right now so we'll have to suspend new signups (old users should be 👌 )
The case against
Paolo Rosson: Jev counts letters at 70% on words, 168 of 168 on lists
Paolo Rosson asked Jev to count the r's in strawberry and got 47% on three and 47% on two, a coin flip. Across 168 test words he measured 70% accuracy, with doubled letters undercounted; feeding the same question as a character list instead of a word got 168 of 168 at 260ms. Diogo Almeida's reply did not explain the gap.
then I gave it the letters instead of the word: ["s","t","r","a","w","b","e","r","r","y"] 168/168. same model, same question, 260ms