Independent developer guide
Jev: the decision model that doesn't write a word
TypeSafe AI launched Jev on September 15, 2026 as the first "System One model". It answers typed questions about a state and hands your code a structured decision — no prose, no JSON parsing, no retries.
What is Jev?
Every LLM you use today is what TypeSafe would call a System Two model: it thinks out loud, one token at a time, and you get a string back. That is wonderful for writing and terrible for deciding. When you only need to know which of these five tools should run next, making a chat model produce JSON and then parsing it is slow, expensive and fragile.
Jev attacks that gap. It gives up string generation entirely in exchange for structured output. You send one state (unstructured text, or your program state) plus any number of typed questions, and you get every answer back in a single call, each with a probability distribution and a confidence value.
The three primitives
Everything Jev does is built from three question types. Each question is evaluated in parallel and in isolation against the same state, so adding more questions barely changes latency.
Choice
“Choose an option from a list.”
You supply a state and a fixed set of options. Jev returns which option fits, a probability for every option, and a confidence value — so your code can branch on the winner or fall back when confidence is low.
Returns: choice, probabilities, confidence
Score
“Score the state on a rubric.”
You define ordered descriptive levels (a rubric). Jev returns a score against that rubric plus the probability distribution across levels, which you can sort or threshold on.
Returns: score, probabilities, confidence
Noul
“Is this statement true?”
A single yes/no question that returns the probability of "yes" rather than a hard boolean — letting you pick your own threshold instead of trusting a coin flip.
Returns: a noul value in the 0–1 range
Full request and response shapes are on the API reference page.
Jev at a glance
- Vendor
- TypeSafe AI
- Model class
- System One model
- Launched
- September 15, 2026 (out of stealth, $40M seed round led by DCVC)
- Model routes
jev-latestjev-1.13- Input price
- $0.042 per 1M input tokens
- Output price
- Free (no output tokens are generated)
- Latency
- 70–500ms end-to-end
- Context
- ~32,000 tokens per request
- Modalities
- Text only — no image or audio input at launch
- SDKs
- Python, JavaScript
- Open weights
- No — closed managed API only
When Jev is the right tool
Reach for Jev when the answer is a decision, not a sentence — and especially when that decision sits on a hot path where a full LLM round-trip hurts:
- Picking which tool or agent handles the next step
- Routing a request to the right model, queue or handler
- Classifying intent, topic, language or priority
- Grading output against a rubric before it reaches a user
- Boolean guardrails: is this on-topic, is this complete, is this safe to send
And skip it when you need prose, code, summaries or a conversation — Jev simply does not produce them. Most real systems end up using both: Jev decides, an LLM writes.
Frequently asked questions
What is Jev?
Jev is TypeSafe AI's flagship model and the first of a class they call System One models. Instead of generating text token by token, it evaluates typed questions against a state and returns structured decisions — a chosen option, a score, or a probability — that your code can branch on directly.
Is Jev open source?
No. Jev is delivered exclusively as a closed, managed API with no open weights and no self-hosting or VPC option. Several community projects reimplement the interface with open models, but they are not Jev.
Can Jev hallucinate?
Jev does not generate free-form strings — it returns a choice, a score, or a probability from a set you define. Because the output space is constrained to options you supply, it cannot invent an answer outside that set. It can still be wrong or over-confident, which is why every answer ships with a confidence value.
How much does Jev cost?
$0.042 per 1M input tokens, and output is free because no output tokens are generated.
Is Jev a replacement for GPT or Claude?
No. Jev does not write text, summarise, or converse. It replaces the decision steps inside your pipeline — routing, classification, tool selection, grading — that you were previously paying a chat model to do slowly and parse unreliably.
jev-agent.com is independent and unofficial. Authoritative docs live at docs.typesafe.ai.