Jev
The first System One model: unstructured state in, typed probabilistic decisions out — in milliseconds.
Jev is TypeSafe AI’s first System One model, built for the fast, structured decisions inside software rather than for chat. You send a state plus typed questions — Choice (pick from a list), Score (rate against a rubric) or Noul (a true/false-style judgment) — and get back schema-defined answers with calibrated probabilities and a confidence value. Questions are evaluated in parallel, so adding more barely changes latency. It uses a new architecture with a parallel sampler and is trained with Reinforcement Learning for Calibrated Decisions (RLCD). Because it never generates free text, its outputs cannot fall outside the schema. TypeSafe reports 70–500 ms end-to-end latency and large speed/cost gains over frontier LLMs on its own workflow evals (vendor-reported; early access via waitlist).
- Provider
- TypeSafe AI
- Type
- Decision model
- Released
- Oct 9, 2026
- Price
- $0.042 input / Free output per 1M tokens
- Input
- text
- Output
- decision
- Open weights
- No
Best for
- Extraction & classification
- Real-time / low latency
- High volume / low cost
- Agents
- Enterprise
Strengths
- Typed outputs that cannot mismatch the schema
- Calibrated probabilities on every decision
- Millisecond-scale latency (vendor-reported)
- Input-only pricing — output is free
Tools that use it
- System One adapter — TypeSafe AI. Turn LLM calls into typed, System One-style decisions.