Author(s): Shobhit Agarwal

Originally published on Towards AI.

Why a model that can’t write a single sentence might be the smartest AI release of 2026

Jev is a new AI model from TypeSafe AI, a startup founded by Diogo Almeida (a former OpenAI researcher who worked on RLHF/InstructGPT), which came out of stealth on September 15, 2026 with $40M in funding led by DCVC. It’s the first public release in TypeSafe’s “System One” model class, named after economist William Stanley Jevons.

After the intro, the article explains the core idea behind Jev: instead of generating tokens as chat LLMs do, Jev takes application state plus structured, typed questions and returns probabilistic decisions in a single parallel pass. It describes Jev’s three question primitives (Choice, Score, and Noul), the RLCD training approach that calibrates confidence as a real statistical signal, and why this design can reduce latency, cost, and schema-related failures compared to frontier models—while noting important caveats about self-run benchmarking and consensus-based ground truth. It then shows practical integration steps (SDK install, calling a single endpoint, and branching in code using typed outputs), lists common use cases like ticket triage, routing, retrieval filtering, and high-frequency decision loops, and highlights where Jev struggles (no “thinking” in natural language, weak at counting and dates, no world knowledge/retrieval, and susceptibility to adversarially crafted state). The piece concludes by framing Jev as a reminder that the smartest AI progress may be building systems that waste fewer words and better match structured decision needs inside real products.

Read the full blog for free on Medium.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI