RELEASE · MODELS · #1347
TypeSafe AI launches Jev, a decision-only model returning typed probabilities
TypeSafe AI released Jev, the first of its "System One" models: a decision-only model that returns typed, probabilistic answers (Choice, Score, Noul) instead of natural language, evaluating multiple typed questions in parallel and emitting probability distributions with confidence values. Jev has a 32,000-token context window, quoted input cost of $0.042 per million tokens (output free), end-to-end latency of 70–500 ms, uses a training method called Reinforcement Learning for Calibrated Decisions, and saw rapid early integrations from Vercel, Netlify and LangChain.
KEY POINTS
- TypeSafe AI released Jev, the first of its "System One" models: a decision-only model that returns typed, probabilistic answers (Choice, Score, Noul) instead of natural language, evaluating multiple typed questions in parallel and emitting probability distributions with confidence values.
- Jev has a 32,000-token context window, quoted input cost of $0.042 per million tokens (output free), end-to-end latency of 70–500 ms, uses a training method called Reinforcement Learning for Calibrated Decisions, and saw rapid early integrations from Vercel, Netlify and LangChain.
- Jev represents a new model category that emits machine-readable, calibrated probabilities for direct programmatic decisioning, which can reduce cost and latency and change how systems route and act on AI outputs while introducing new reliability trade-offs.
WHY IT MATTERS
Jev represents a new model category that emits machine-readable, calibrated probabilities for direct programmatic decisioning, which can reduce cost and latency and change how systems route and act on AI outputs while introducing new reliability trade-offs.