OpenAI's Decisions API hits market three weeks after Jev

Almeida's idea predates both products. While still at OpenAI, he wrote a document outlining models optimized to work with software and showed it to CEO Sam Altman, who urged him to set aside his other work and pursue it. OpenAI never rolled out its own classification engine until Jev showed up and caught fire among developers.

The gap Almeida targeted is real: simple machine learning classifiers have handled tasks like spam filtering for decades, but they are inflexible and require technical know-how. LLMs classify with more nuance and can be redirected in plain language, yet cost more and run slower — a problem that compounds across thousands of daily calls. Jev is named for economist William Stanley Jevons, whose paradox — cheaper steam power leading to more coal use — Almeida applies to AI, betting lower costs will make many more tasks worth automating.

Developers report mixed experiences. Niels Mouthaan was impressed by Jev's speed. Ashutosh Mathore reconsidered where he was using LLMs 'out of habit.' But Aaron Roy found Jev let important emails slip until he spelled out specific criteria, and Paolo Rosson estimated Jev's calls at seven cents per 1,000 changes. DCVC's James Hardiman argues TypeSafe's two years of reliability work gives it an edge over 'fast follow clones,' though whether Jev can hold its ground against OpenAI remains uncertain.