A new kind of AI model from a ChatGPT inventor is thrilling developers — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- TypeSafe AI, founded two years ago by ex-OpenAI researcher Diogo Almeida, released Jev, a transformer-based model that outputs "calibrated decisions" — probabilities — rather than text, which the company says makes hallucination impossible since users define outputs in advance.
- Demand for Jev briefly overwhelmed the company's API capacity at launch; the model meters input tokens by the billion and offers free output tokens, positioning it as a low-cost automation primitive.
- Vercel replaced OpenAI's ChatGPT Luna 5.6 with Jev for a safety-classifier workload and reported results 5 to 18 times faster alongside greater accuracy, per software engineer Pranit Sharma.
- Bryo AI CTO Nikhil Mudholkar pitted Jev against Google's Gemini for business-email classification: Gemini was slightly more accurate but cost 10 to 20 times more, with Mudholkar crediting Jev's real confidence scores as "ideal for automating workflows."
- Jev is trained exclusively on synthetic data using a technique Almeida calls "reinforcement learning from calibrated decisions," and is named after 19th-century economist William Stanley Jevons, whose paradox holds that falling cost drives broader use.
- Almeida drew a sharp line against frontier labs: "the main product of frontier labs is fear or hype. I would like our main product to be intelligence," telling TechCrunch TypeSafe is "not a lab in the sense of, you know, like bet on infinite wealth, or a religion, or building God in a data center."
Why it matters: For teams running classifiers, routers, and LLM-monitoring jobs in production, Jev demonstrates that a non-language model can beat frontier APIs on cost and latency — Vercel cut safety-classifier latency 5-18x versus OpenAI, and Bryo AI found Jev 10-20x cheaper than Gemini per task. Almeida's Jevons framing — that cheaper intelligence gets used more — points to automation workloads quietly migrating off frontier-model APIs toward purpose-built alternatives.
Ask SkimNews
