No Chatter, Just Judgment: Jev Model Wins 13% of Paid Teams in 24 Hours

创业邦 · wechat · 2026-09-26

Jev, a "System One Model" from TypeSafe AI, generates no text — it only returns choices, scores, or calibrated probabilities from predefined options. Within 24 hours of launching on Vercel's AI Gateway, nearly 13% of paid teams adopted it, more than double the rate of comparable model launches, with 140k developers joining the beta in 36 hours.

Background

Founded in 2024 by ex-OpenAI researcher Diogo Almeida, a lead author of InstructGPT and foundational RLHF contributor, alongside Erik Gafni and Sasha Sheng
Raised a $40M seed round led by DCVC; valuation reportedly $200M

Technical approach

API supports only Choice (up to 255 options), Score, and Noul (yes/no probability); no text generation or reasoning chains
End-to-end latency of 70-500ms, up to 193x faster than frontier LLMs; input at $0.042/M tokens, output free, $0.0004 per structured decision
Trained with RLCD so stated confidence matches actual accuracy — third-party reviewers call it the only model returning true probabilities
Retains zero-shot generalization without retraining

Market

Vercel engineers replaced an existing classifier with Jev, gaining 5-18x speed and higher accuracy
A surprisingly hot use case: WeChat community-management plugins for moderation triage
Open-source reproductions appeared within three days, suggesting the moat lies in ecosystem and brand rather than architecture

The thesis: Jev complements rather than replaces LLMs — expensive models do slow planning, while high-frequency atomic judgments go to cheap, reliable decision models.

Original post →

No Chatter, Just Judgment: Jev Model Wins 13% of Paid Teams in 24 Hours

More from Venture