TypeSafe's Jev never outputs a word — typed decisions with confidence scores, a listed $42 per billion input tokens, and a launch-week API that demand briefly broke

The company's framing, from its September 15 post: Jev is 'a new class of frontier models built to make fast, structured decisions that software can use directly', 'available today in early access', and 'While Jev gives up string generation, it's optimized for structured outputs and can't hallucinate.' The pitch line: 'Think of Jev as a frontier-intelligence function call: unstructured state in, typed probabilistic decisions out.' TechCrunch's translation: 'a new transformer-based model, Jev, that is not a large language model (LLM). It doesn't output text, but instead produces probabilities, or what the company calls "calibrated decisions."' Who built it: Diogo Almeida, whom TechCrunch credits as an OpenAI researcher who 'helped build the chatbot and then invent reinforcement learning from human feedback (RLHF)', and who told the outlet 'We have lightning in a bottle, and yet it is not useful' about his ChatGPT-era work. The money, per the company's own table: 'Input tokens: $0.042 / MTok ($42 per billion tokens).' and 'Output tokens: FREE (too cheap to meter)' — with the company's own caveat 'We can't prove it isn't subsidized' about its own pricing. The speed claims are the company's: '70ms-500ms' end-to-end, '40x-200x faster' on System One-shaped queries, the homepage's '193.6x Faster, 444.6x Cheaper.' footnoted '*based on workflows for System One tasks' — and the blog's own concession that 'we expect that these are on the higher end of real world gains' — plus '238x Lower input price than Claude Fable 5.1'. The independent beats are TechCrunch's: 'the company briefly lost the ability to serve users from its API because demand was so high'; Vercel's Pranit Sharma got results 'five to 18 times more quickly and with greater accuracy' after replacing OpenAI's ChatGPT Luna 5.6 with Jev on a safety-command classifier; Bryo AI CTO Nikhil Mudholkar found Gemini slightly more accurate but '10 to 20 times more expensive' for classifying business emails, and called Jev's confidence scores the real difference: 'it is the only one that hands back a real probability which makes it ideal for automating workflows!!' The usage frame, from Armin Ronacher (CTO of Earendil, builders of the open source model harness Pi): 'it delegates the hallucination problem a little bit to the user' — at 50% 'maybe this is a coin toss, and I disregard it. But if it's 95%, sure, then I can do something with it.' The secrecy: 'Almeida is tight-lipped about the model's architecture, which outside observers suspect is built on top of an open-weight LLM', Almeida says it is 'trained exclusively on synthetic data using a technique he calls "reinforcement learning from calibrated decisions."' — the company's own rendering of that name is 'Reinforcement Learning for Calibrated Decisions (RLCD)'.