TypeSafe AI has released Jev, a transformer-based "System One" model that returns typed decisions with probabilities rather than generating text. The system accepts state and typed questions via a single API endpoint, supporting Choice, Score, and Noul primitives to enable code branching based on calibrated confidence.
- Jev uses Reinforcement Learning for Calibrated Decisions (RLCD) to output confidence values derived from probability distribution shapes.
- Input costs $42 per billion tokens, with output tokens being free.
- TypeSafe claims 193.6x speed and 444.6x cost advantages over GPT-5.6 Terra in its own workflow evaluations.
- The model is available as a hosted API in early access behind a waitlist; weights and parameter counts are not disclosed.
The release aims to move AI from human-facing chat to software integration by providing deterministic, branchable outputs with explicit uncertainty metrics.