Jev: New frontier model 40-400x cheaper and 20-200x faster
8 hours ago
- TypeSafe AI released System One Models, a new class of frontier models optimized for fast, structured decisions.
- The first model, Jev, provides type-safe, calibrated probabilistic outputs without hallucination, using RLCD training and parallel sampling.
- Jev is 40-200x faster and cheaper than traditional LLMs, with end-to-end response times of 70-500ms and free output tokens.
- It is designed for automation tasks such as classification, routing, scoring, and extraction within software workflows.
- Demos include real-time Doom gameplay and Wikiracing, demonstrating intelligence-per-second and compound benefits of no hallucination.
- Early access is available now, and the company seeks developer feedback for further use cases.