Updated

Jev (early access)

TypeSafe AI released Jev, its first public System One model, in early access. It evaluates supplied state and returns typed decisions with probabilities instead of generating free-form text.

By No Code MBA · Facts checked

Provider
TypeSafe AI
Announcement date
Status at last check
Limited access or preview
Model family
System One
Capabilities
Text

What changed

Jev is TypeSafe AI's first System One model, built to return predefined typed values with calibrated probabilities rather than free-form text. It takes text and structured program state as input and is trained with a method the company calls Reinforcement Learning for Calibrated Decisions. TypeSafe AI reports 70 to 500 ms end-to-end latency, 40x to 200x faster than frontier LLMs on suitable queries, and 0% schema errors by design.

Access and availability

Early access through TypeSafe AI's API console and docs, with developers being admitted from a waitlist. No open weights are mentioned.

Official access documentation

Pricing

$0.042 per million input tokens; output tokens free

Official pricing

What you can build

Try classifying support tickets or routing requests using predefined choices, with a review path for uncertain decisions.

Read our Jev AI guide

Limitations

Jev cannot generate strings, and direct choices are limited to a cardinality of 255, with a slower two-stage path above that. TypeSafe AI notes its evaluation references average GPT-6 Astra and Fable 5.1 answers and some workflows were built in-house, so results may be biased; model size and training data are not disclosed.

Sources and release history