Chatting With Jev as a Regular LLM: A 3 tok/s Caveman-Style Experience

_rchaves_ · x · 2026-09-18

A user tried using TypeSafe's judgment model Jev as a regular chat model. Verdict: not a great idea — Jev generates at only 3 tok/s vs 150 tok/s for normal LLMs, has no caching so it ends up more expensive, and isn't RL'd for conversation. But he just wanted to talk to the model and learn how it "feels," even in caveman style, and finds it exciting to have a primitive that lets you kind of rebuild an LLM from scratch.

Related event: Developer Forces Judgment Model Jev to Talk, at 3 tok/s(3 posts)→

Original post →

More from Fun

Fun channel →