ChatGPT co-inventor launches Jev model claiming 20-200x speed and 40-400x cost cuts
multiply_matrix · x · 2026-09-18
A ChatGPT co-inventor unveiled Jev, a new frontier model trained with a novel method called RLCD after two years in stealth, claiming 20-200x faster inference, 40-400x lower cost with free output tokens, and 'frontier composable intelligence' optimized for decisions.
The post frames it with a neuroscience analogy: brains need a fast, dumb System 1 to free the slower, smarter System 2 for what matters. Likewise, a model smart enough to control a robotic hand can't also do frontier math if its context window is polluted with vision/control tokens — pairing a powerful model with a faster, still-intelligent one is the way forward.
More from Models
- Jev playground shows inference and roundtrip latency; EU users pay 120ms extra — DanielLockyer · 2026-09-18
- ChatGPT has stopped searching and started guessing: car repair data is all made up — WeissMISFIT · 2026-09-18
- Tesla publishes 12-month FSD Supervised safety report card — yunta_tsai · 2026-09-18
- GPT-6 Astra Cracks a Previously Unsolved 1918 German WWI Radio Message — iamfakhrealam · 2026-09-18
- Matt Holden: structured output is LLMs' real magic, now at 300ms and nearly free — holdenmatt · 2026-09-18
- Wrong Prediction, Right Answer: EMNLP paper shows a 'readout bottleneck' hides LLM reasoning — 机器之心 · 2026-09-18