TypeSafe AI has released Jev, a model that doesn't produce text but instead delivers typed, probabilistic decisions that software can execute immediately. The San Francisco lab, founded by former OpenAI researcher Diogo Almeida, calls it the first System One Model. Within a day of launch, Jev reached nearly 13% of Vercel's paid teams, double the adoption rate of the GPT-5.6 family, according to a report published by InfoQ in October 2026.
Jev processes questions in a single parallel pass and delivers Choice, Score, and Noul answers that include probability distributions and confidence values, allowing software to act when confidence exceeds a threshold and escalate when it falls short. Input pricing stands at $0.042 per million tokens with free output, the context window spans 32,000 tokens, and latency ranges from 70 to 500 milliseconds end to end. An analysis of 12,759 launch tweets by OpenChamber found that users reported a median speedup of 7x, cost savings at a median of 30x, and latency at a median of 76 milliseconds with an upper quartile of 270 milliseconds. Vercel engineer Pranit Sharma reported that a safety classifier ran five to 18 times faster than the language model it replaced, while Bryo AI CTO Nikhil Mudholkar found Gemini slightly more accurate on email classification but 10 to 20 times more expensive, and valued Jev as the only option returning a genuine probability.
Armin Ronacher, CTO of Earendil, told TechCrunch the design "delegates the hallucination problem a little bit to the user," who must decide whether a 50% probability justifies action. On Hacker News, one developer congratulated the team but noted that while Jev can't emit an invalid type, it can still produce a completely wrong valid value, adding that "can't hallucinate" seems inaccurate. The model's documentation warns of unreliable counting, arithmetic, and date comparison, plus accuracy loss on large noisy state, and recommends keeping mathematical operations in code rather than delegating them to the model.
The speed advantage comes from Jev's narrow design: it evaluates structured questions against a state and returns typed answers, not open-ended text generation. Training relies on a method TypeSafe calls Reinforcement Learning for Calibrated Decisions. Vercel added Jev to AI Gateway on day two, Netlify followed, LangChain shipped a TypeSafeClassifier integration with model routing and an AutoMode middleware that screens tool calls before execution, and five independent Elixir clients appeared within days. The report advises teams replacing language model classifiers to start with the jev-1.13 jaggedness page, pin a specific version such as jev-1.13.0 rather than moving aliases, and use the System One adapter to benchmark existing models against the same schema. One developer on Reddit called the model "absolutely insane" for agent work at 200 to 300 milliseconds latency, while an early access user on Hacker News called it "really neat" but cautioned that its out-of-distribution behavior will differ from a language model. TypeSafe AI was founded by Diogo Almeida, Erik Gafni, and Sasha Sheng; Almeida co-invented RLHF and contributed to the research behind ChatGPT. The choice between returning a probability distribution and generating explanatory text represents a fundamental architectural tradeoff, one that enterprises will increasingly confront as they separate routing decisions from reasoning tasks.