Jev is built to make software decisions, not chat
Jev returns probability scores for decisions like fraud risk, review or approval in 70-500ms, aimed at high-volume small decisions such as risk checks, recommendations and agents.
JEV IS NOT TRYING TO BE ANOTHER CHATGPT ✨ TypeSafe has introduced Jev, an AI model built for something much simpler: making decisions for software. Instead of generating a long answer, it can give the software a result like: → Fraud risk: 94% → Review: 97% → Approve: 3% This can be useful for fraud detection, recommendations, AI agents, risk checks, data classification, games and other things where software needs to make lots of small decisions quickly. The interesting part is the speed. Normal AI models generate text one piece at a time. Jev takes a different approach and is designed to make these decisions much faster. TypeSafe says Jev can respond in 70–500ms and costs $0.042 per 1 million input tokens. Output tokens are currently free. In TypeSafe’s own testing, Jev was up to 193.6× faster and 444.6× cheaper than the other models they tested. These are TypeSafe’s numbers, so independent testing will be important. Jev also gives a confidence score with its decision. So software can handle high-confidence cases automatically and send uncertain ones to a human. TypeSafe calls this approach System One Models and uses its RLCD training method to improve both decisions and confidence. ChatGPT is built to talk to people. Jev is built to make decisions for software.