Diogo Almeida, an OpenAI researcher who helped build ChatGPT and invent reinforcement learning from human feedback (RLHF), left OpenAI two years ago to start TypeSafe AI. This week the company released Jev, a transformer-based model that is not a large language model — it outputs calibrated probability decisions instead of text.
Eschewing language makes the model incredibly cheap and fast, and because users define the outputs in advance, it cannot hallucinate. Its output tokens are free, and input tokens are metered by the billion, not the million.
Developers are taking great interest; the company briefly lost the ability to serve users from its API because demand was so high. Vercel replaced OpenAI’s Luna with Jev for command safety classification and got results 5-18x faster with greater accuracy. Bryo AI found Jev 10-20x cheaper than Gemini for email classification, with useful confidence scores.
Jev can augment LLMs as a cheap agent monitor for jailbreaks and model routing. Named after economist William Stanley Jevons, the model is trained exclusively on synthetic data using “reinforcement learning from calibrated decisions.”
TypeSafe refers to Jev as a “System One model” focused on intuition rather than reasoning. Architecture details are undisclosed; outside observers suspect it may be built on an open-weight LLM base.