JevStation

JevStation

JevStation is a workspace for running TypeSafe's decision model Jev: submit a piece of state along with typed questions and get structured answers with probability distributions and confidence that your code can branch on directly.

paid
Last Updated 10/13/2026

About JevStation

JevStation is a workspace for running Jev, TypeSafe's System One decision model. You submit a piece of state (a ticket, a review, a tool call, or a model answer) along with questions, and get typed answers: each answer is a value your code can branch on directly, with a full probability distribution and confidence, with no need to parse generated text. Three question types are supported: Choice (select from options you define), Score (rate against an ordered scale), and Noul (judge whether a statement is true or false, returning 0–1). Questions work best when each targets one clear, well-scoped specific judgment; multiple questions evaluate the same state in parallel, so adding questions adds almost no latency. JevStation also offers a browser Playground, savable evaluation threads, reusable question sets, and API keys. Billing is per evaluation rather than per token: a standard evaluation costs 1 credit (state up to 8,000 characters and up to 5 questions), and larger ones cost 3 credits; failed evaluations cost 0 credits.

Best for

Developers who need to consume structured judgment results in their own code
Engineering teams automating judgments on tickets, code reviews, or model outputs
Users who want to try things in the browser first before wiring up the API

Best use cases

Wire programmatic, consumable judgment results into your code — for example, branching on typed values from Choice / Score / Noul

Make structured judgments on tickets, code reviews, tool calls, or model outputs

Run combined evaluations of the same input with multiple parallel questions

Quickly try things out in the browser Playground without writing any code

Compare before choosing

According to the official comparison page: per token, Jev is the cheapest of the three — a typical evaluation costs about $0.000020 on Jev, about $0.000033 on Clef-flash, and about $0.000089 on Clef (list prices).

According to the official comparison page: for longer inputs (long threads, documents, transcripts) both can read them; Jev is cheaper per token, but Clef has a larger context window.

According to the official comparison page: when you need to judge images, choose Clef or Clef-flash — Jev does not accept images; for short inputs, high concurrency, or latency-sensitive use cases, start with Clef-flash; for fixed intent lists and tool lookup scenarios, Cloudflare's own data favors Clef.

Features

Typed answers

Choice, Score, and Noul questions return values your code can branch on directly, with full probability distributions and confidence — no text parsing needed.

Atomic questions, composed in code

Each question targets one clear, well-scoped specific judgment; multiple questions evaluate the same state in parallel, so adding questions adds almost no latency.

Online Playground and API

Run evaluations right in the browser with no code; after registering an account, create API keys to integrate into your code.

Evaluation workspace

Save evaluation threads and reuse question sets; failed evaluations automatically refund credits.

One-time credit billing

No subscription, no automatic renewal; run 3 preset evaluations per day without registering, and get 50 credits when you sign up.

Pros

  • Answers are typed values your code can branch on directly, with probability distributions and confidence — no parsing from text
  • Three question types (Choice / Score / Noul) cover multiple choice, rating, and true/false judgments
  • Evaluations either return a complete result or fail cleanly; failures cost no credits and are automatically refunded
  • No subscription or long-term commitment — billed per evaluation, with 50 credits on sign-up and 3 free tries per day without registering
  • Includes a browser Playground, savable evaluation threads, reusable question sets, and API keys

Cons

  • Jev does not accept image input; use Clef or Clef-flash when you need to judge images (per the official comparison page).

Pricing Plans

5,000 credits

$9.90one-time
  • One-time purchase, no recurring charge

12,000 credits

$15.90one-time
  • One-time purchase, no recurring charge

* Pricing information is for reference only. Please refer to the latest announcement on JevStation's official website.

FAQ

No. Credit packs are one-time purchases — no subscription, no automatic renewal, and nothing to cancel.
A standard evaluation costs 1 credit; it costs 3 credits when the input state exceeds 8,000 characters or there are more than 5 questions. Failed evaluations cost no credits.
Yes. Unregistered users can run 3 preset evaluations per day; after registering you get 50 credits and can create API keys.
According to the official comparison page: for longer inputs, Jev is cheaper per token but Clef has a larger context; for judging images, choose Clef or Clef-flash; for short inputs, high concurrency, or latency-sensitive use cases, start with Clef-flash; for fixed intent lists and tool lookup scenarios, Cloudflare's own data favors Clef.

Tool Information

DeveloperJevStation
Release Date10/13/2026
Pricingpaid