Jev Model
A decision model for software: send a state and typed questions, and receive structured judgments with probabilities for classification, routing, scoring, and safety checks.
About Jev Model
Jev Model is a decision-model service for software: you send a context state and a set of typed questions, and it returns structured answers along with probabilities you can use in your own workflows. Unlike traditional large language models, which mainly generate text for humans to read, it focuses on letting software directly consume judgment results. It currently accepts text, JSON objects, and text arrays as input; image, audio, and video are not yet supported, and accuracy for non-English input requires separate validation.
Best for
Best use cases
Classify tickets, orders, or policy content and feed the probability results into code logic
Make routing decisions for messages or records, replacing manual reading and judgment with structured output
Score or run safety checks on input states, using typed answers and confidence levels as downstream evidence
Validate a real workflow in the online playground first, then integrate it into your product via the API
Compare before choosing
The official comparison page states: Jev is stronger on calibration (82.7 vs 65.4) and hard-tier accuracy (74.1% vs 69.5%), and offers real pricing, rate limits, and a pinnable version; djev is faster on speed (91.4 vs 83.3), and natively supports image input, image options, and real-time camera sampling, free during the preview period.
The official comparison page states: djev supports self-hosting (roughly requiring an NVIDIA B200, Linux, CUDA 13, and BF16), while Jev Model does not offer self-hosting.
The official comparison page states: their intelligence scores are close (85.7 vs 82.7), simple decisions rarely show a meaningful gap, and the real differences lie in calibration, input modalities, and the amount of infrastructure operations work.
Features
Structured decision output
Send a piece of context and explicit questions, and receive structured answers with type annotations, probabilities, and confidence scores.
Multiple question types
Supports three question types: choice, score, and noul; the Pro plan also supports asking multiple questions in parallel per request.
Low-latency evaluation
Typical decision latency is about 120ms, with up to 8 questions answered per request.
HTTP API
Provides a POST /v1/systemone endpoint — send a request with a Bearer API key to evaluate a state and receive structured answers question by question.
Online playground and key management
Try it directly on the web and manage your keys through the API key management page.
Pros
- Returns typed decisions and probabilities that software can consume directly, rather than free-form text
- The official website offers free online usage with no traffic limits
- Typical decision latency is about 120ms, with 8 questions answered per request
- Provides a full HTTP API and documentation, making it easy to integrate into production workflows
Cons
- Currently only accepts text, JSON objects, and text arrays; image, audio, and video inputs are not yet supported.
- The official documentation notes that accuracy for non-English input requires separate validation.