DiffusionGemma becomes an open, self-hostable decision model on vLLM amid Jev's viral 'System One' moment

rseroter · x · 2026-09-29

A Red Hat Developer post shows how to run DiffusionGemma on vLLM and Red Hat AI to build an open, self-hostable decision engine. Context: Jev from TypeSafe AI (founded by ex-OpenAI researcher Diogo Almeida) recently went viral — a model that doesn't chat but returns typed, probabilistic decisions via three question types (Choice, Score, Noul), inspired by Kahneman's 'System One' thinking. Enterprises already run small models as zero-shot classifiers; Jev's contribution is a model and API purpose-built for the role: low latency, low cost, guaranteed structure, probabilities as normal output. DiffusionGemma was already a Red Hat AI validated model, and the vLLM community turned it into a self-hostable alternative.

Original post →

More from coding & agent

coding & agent channel →