What is Jev? See one support example
Jev is a new AI model from TypeSafe. See how it chooses from defined answers in one support example.
Jev in one sentence
Jev is a new AI model from TypeSafe, founded by Diogo Almeida, an early contributor to core ChatGPT research. It helps software judge, choose, and score instead of writing long answers.
TypeSafe calls Jev a System One model. You can set that term aside for now. Routing a support message, picking a relevant document, or choosing a game action can all become a choice among known options.
Follow one support message
A customer writes: “My parcel has not arrived. Can you check?”
- Your application gets the message. If order details are needed, it supplies authorized data.
- Jev chooses. Among delivery, refund, payment, and other, it returns how likely each option is.
- Your application acts. It routes the ticket to the delivery team, or asks a person to review an uncertain result.
Jev does not open an inbox, write the reply, or issue a refund by itself. Its job is the choice in the middle.
What three kinds of questions can it answer?
| Plain-language task | Everyday example | Official name |
|---|---|---|
| Yes or no | Should a person review this message? | Noul |
| Pick one | Is this about delivery, refunds, or payment? | Choice |
| Rate it | Is this ticket low, medium, or high priority? | Score |
You define the possible answers so software can use the result. A valid answer can still be the wrong answer.
Why is it getting attention?
Software makes many small, repeated decisions. TypeSafe built Jev for these tasks and demonstrates fast responses and low input prices. It has also demonstrated decisions in Doom using prepared game state. Actual speed, cost, and accuracy depend on your task, inputs, and provider. The large performance multipliers on TypeSafe's site come from specific workflows, not a guarantee for your application.
What should it not do?
- Write or chat freely: use a generative model for emails, articles, code, or open-ended answers.
- Calculate exactly: use ordinary code for money, counts, and dates.
- Read raw images or sound: Jev currently takes text; the game demo used structured state.
Test your own language and task before relying on the result. Keep permission checks and human review for sensitive actions.
Next step
Pick one small decision and ask: “Can I write down the possible answers?” If yes, read how to choose the question type.
Sources and verification boundary
Reviewed September 19, 2026. Documentation-based guidance, not a live API or business-performance test.
TypeSafe · Introduction →TypeSafe · How to build →TypeSafe · Jev 1.13 limitations →TypeSafe · Models, pricing and limits →TypeSafe · Founders and team →OpenAI · Introducing ChatGPT →