Jev AI
A typed decision model from TypeSafe: send application state and get back a choice, score, or yes/no probability that your code can branch on directly, billed on input tokens.
About Jev AI
Jev is a decision model from TypeSafe (currently publicly aliased as jev-1.13.0), officially positioned as a "typed decision model", not a chatbot. How it works: you send application state (for example, a ticket description) to the hosted API, and the model returns typed results your code can branch on directly — a choice that picks one value from a known set (routing), a score on an ordered scale (ranking), or a noul yes/no probability (gating). The official website explicitly states that the model was not trained to generate text, so it is not suited to open-ended language tasks like writing emails or explaining error messages. The officially recommended pattern is hybrid use: let Jev decide the queue, the next model, or whether a tool call is allowed, while GPT generates replies, summaries, or patches, and escalate to GPT or a human when Jev is uncertain. The API officially opened on September 21, 2026 by removing the waitlist — just log in and create a key to start using it.
Best for
Best use cases
Automatically classifying support tickets into a known set of topics (such as billing)
Scoring and ranking tickets or leads by urgency, risk, or review priority
Deciding in a call chain whether a given tool call is allowed (gating)
Serving as a routing layer to choose the next model to execute, handing language tasks like conversation and summarization to GPT
Compare before choosing
The official comparison page states: Jev excels at bounded labels, scores, and yes/no probabilities; GPT excels at language, tools, and long reasoning; the recommended combination is "Jev decides, GPT expresses".
The official comparison page states: Jev is not stronger at writing, coding, or open-ended reasoning; when you already know the allowed answers and need cheap typed branching, Jev is the better fit, while GPT is better when the output is language.
The official comparison page states: GPT's input and output prices are higher, with output usually the most expensive part; the vendor's latency chart also shows Jev leading on System One–type tasks, but you should still measure p95 latency from your own region.
Features
choice: routing
Pick one value from a known set; officially recommended for support ticket topic classification, model routing, and intent detection.
score: ranking
Place an input on an ordered scale; useful for urgency, lead quality, risk, or review priority.
noul: gating
Returns a yes/no probability for code to read and branch on directly.
Public API
The hosted API opened on September 21, 2026 by removing the waitlist; log in and create a key to start calling it.
Pros
- Outputs three typed result types — choice / score / noul — so code can branch directly on result.choice or result.noul without parsing keywords from sentences
- Billed on input tokens with output free of charge
- The API is already open (no waitlist) — just log in and create a key to use it
Cons
- The official pages explicitly state that jev-1.13 was not trained to generate text and is unsuitable for language-output tasks such as drafting emails or explaining error messages
- The official pricing estimate page notes that its estimates exclude the costs of retries, gateways, and fallback LLMs, so actual costs must be measured yourself
Pricing Plans
Pay as you go
- Input is billed at $0.042 / 1M input tokens
- No output charges
* Pricing information is for reference only. Please refer to the latest announcement on Jev AI's official website.