Jev, from a ChatGPT researcher, claims 200x faster and 400x cheaper decisions than LLMs
bibryam · x · 2026-09-22
- TypeSafe AI, founded by ChatGPT researcher Diogo Almeida, launched Jev, a "System One" model for structured decisions, claiming 200x faster and 400x cheaper judgments than LLMs in its own workflow evals (acknowledged as best-case numbers).
- The insight: many LLM calls only need a classification, choice, or score. Jev takes context plus questions with defined answer types and returns probabilities, option distributions, or scores — your code decides what happens next, no generated text to wait for.
- The post includes a minimal Vercel AI SDK 7+ example (experimentalevaluate), with boolean, choice, and score question types.
- It also surveys eight early projects, from browser agents to context filtering, showing where low-latency judgments can replace LLM calls.
More from coding & agent
- Claude as a One-Person Company: A Full Org Chart of AI Skills — Shruti_0810 · 2026-09-22
- Xiaomi's Luo Fuli Recaps MiMo-V2.6 RL Run: 25K Agent Trajectories Per Step — bookwormengr · 2026-09-22
- Worker Previews auto-provisions isolated Durable Objects and containers per preview — dinasaur_404 · 2026-09-22
- Worker Previews config model: branch-level var/secret overrides without touching prod — dinasaur_404 · 2026-09-22
- Cloudflare launches Worker Previews: production-like environments for every agent change — dinasaur_404 · 2026-09-22
- 2019 Essay on Senior Developers Still Rings True: It's About Judgment, Not Code — tlakomy · 2026-09-22