Jev, the model that never writes a word: 13% of paid teams onboard in first day, 70ms latency

创业邦 · wechat · 2026-09-22

Silicon Valley devs are buzzing about Jev from TypeSafe AI — a model that can't chat, code or write, and only makes judgments: given an input and predefined options, it returns a choice, score or probability. Dubbed a "System One Model," it was picked up by nearly 13% of paid teams on Vercel AI Gateway within 24 hours (2x+ any prior model launch), with 140,000 devs joining the beta in 36 hours. Founded by ex-OpenAI researcher Diogo Almeida, the company emerged from two years of stealth and announced a $40M seed round led by DCVC at a $200M valuation.

Its selling points are speed, price and calibrated confidence: 70–500ms end-to-end latency (vs. 3+ seconds for frontier LLMs, up to 193x faster on official benchmarks), $0.042 per million input tokens with free output, $0.0004 per structured decision. The API supports only three calls: Choice (up to 255 options), Score, and Noul (a 0–1 probability). Trained with RLCD to reward calibrated probabilities, it retains zero-shot generalization but offers no generation, explanation or chain of thought — with no direct competitor on the same technical route yet.

Original post →

Jev, the model that never writes a word: 13% of paid teams onboard in first day, 70ms latency

More from Venture