Stealth model Jev claims 20-200x speedup at 40-400x lower cost; early user runs 5,000 requests for $2

hackgoofer · x · 2026-09-16

What is Jev

A self-described ChatGPT co-inventor (@CompleteSkeptic) launched Jev, a frontier model trained with a new method called RLCD after two years in stealth, claiming:

20-200x faster, 40-400x cheaper (with free output tokens);
Positioned as "composable intelligence optimized for decisions" — a claimed shortcut to AI-driven economic revolution (marketing-heavy, unverified).

Early hands-on

User @MichaelLee04 got access and ran 5,000 requests ($2) across classification, model routing, intent detection and steering.

He sees Jev as a new decision-making primitive between deterministic code and LLM calls — smarter than dumb code, faster and cheaper than LLMs.
He plans to add a Jev call before and after every LLM call in his product, arguing any LLM-powered company could improve quality and tool calling nearly for free.

Original post →

Stealth model Jev claims 20-200x speedup at 40-400x lower cost; early user runs 5,000 requests for $2

More from coding & agent