TypeSafe AI, founded by former OpenAI researcher Diogo Almeida, has unveiled its inaugural AI model, Jev, and secured $40 million in seed funding, valuing the company at roughly $200 million. Jev operates as a System One model, diverging from mainstream large language models (LLMs) that generate text sequentially. Instead, it directly produces structured judgments, probabilities, and confidence scores. By leveraging a parallel sampler and the RLCD training method, the model refines probability calibration. According to TypeSafe AI, in specific structured workflows, Jev can execute tasks nearly 200 times faster than leading LLMs while slashing costs by over 400 times. Moreover, Jev’s mechanisms prevent outputs that deviate from predefined structures or contain type errors, though its judgments may still have inaccuracies, with performance metrics varying by task. Jev is tailored for structured decision-making scenarios, such as classification and routing, aiming to pioneer a new AI division of labor: LLMs handle open-ended text generation, while models like Jev manage high-volume, low-cost, low-latency judgment tasks. However, its real-world value will hinge on its adaptability across diverse use cases.