Jev: The ChatGPT Co-Creator's System One Model Can't Talk
Diogo Almeida’s new model cannot talk. Jev, released in early access on 15 September by TypeSafe AI, returns typed decisions with calibrated confidence: a choice from up to 255 options, a score, or a yes/no probability. It is not a language model, and the man who helped build ChatGPT says that is the point. Jev runs 70 to 500 milliseconds end to end, prices input at $0.042 per million tokens, and…
Jev, a new model developed by Diogo Almeida, does not have the ability to generate language like chat models. Instead, it provides typed decisions with calibrated confidence, offering a choice, a score, or a probabilistic yes/no response. Almeida, a former OpenAI researcher who worked on RLHF and InstructGPT, is behind this model.
The decision-making model is focused on efficiency and accuracy, rather than conversational abilities. It can process inputs at a rate of 70 to 500 milliseconds and costs $0.042 per million tokens, with no charge for output. Jev is designed to make decisions quickly and cost-effectively, making it suitable for use in game loops or real-time decision-making tasks.
The training method, called RLCD (Reinforcement Learning for Calibrated Decisions), aims to train the model for calibrated decisions rather than human preferences, which can lead to overconfidence and other failure modes. Benchmarks show that Jev is significantly faster and cheaper than other models, with an average speed that is 193.6 times faster and 444.6 times cheaper than GPT-6 Astra and Fable 5.1.
While the model's performance is yet to be tested in production, the claim of calibrated decisions and low cost is promising for automation and decision-making tasks.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
