Skip to content
Metatext Decide — typed decisions for your agents, with a probability on every answerGet early access
Metatext
Sign Up

OpenJev

OpenJev on Hugging Face: open-weights decision model for text, web pages and screenshots with up to 52 options per forward pass.

ServicebetaCC-BY-NC-4.0
Open source ↗

Overview

OpenJev (openjev/openjev on Hugging Face) is an open-weights decision model. You describe the decision in plain words at request time with your own labels and get a choice, a yes/no probability or a score, read from the scores at the first output position with fixed calibration settings rather than generated text. The same model can route a support ticket, flag a policy violation, judge whether an answer is grounded, or tell a browser agent which element to act on, the operation to perform, and when the task is finished.

The card reports about 80 ms for short text decisions and about 210 ms for an isolated web decision with about 1,460 prompt tokens and 23 candidates on one H100, up to 52 options per question, and option-order instability of 2.3% versus 18.5% before tuning. It serves the /v1/systemone shape. Weights are CC BY-NC 4.0 (non-commercial) and the helper and server code Apache-2.0, per the list summary. It is separate from SemIf, formerly named OpenJev.

What the decision model decides

TypeQuestionAnswers
ChoiceWhich team should handle this?billing · shipping · technical
Yes / noIs the customer angry?yes · no
ScoreHow urgent is this?can wait · this week · today · right now
ChoiceWhich page element should the browser agent act on next?—

How it works

  1. Serve the model (e.g. with vLLM) behind /v1/systemone.
  2. Send state and questions with labels defined in the request.
  3. Read calibrated probabilities per option.
  4. Act, or escalate on low confidence.

Strengths

  • Labels in the request; no per-task training.
  • Handles DOM, JSON and screenshots for agents.
  • Low option-order sensitivity after tuning.

Caveats

  • Weights are non-commercial (CC BY-NC 4.0).
  • Latency figures are from an H100.

Getting started

Serve the model locally and POST to http://localhost:3000/v1/systemone with model openjev, as in the model card example.

At a glance

Type
Service
Status
beta
Language
Python
Runtime
vLLM, HTTP server
Has code
yes
License
CC-BY-NC-4.0

Use cases and integrations

How curated lists describe it

  • Independent open-weight decision model that scores caller-defined options in one forward pass, with calibration and serving code; weights are CC BY-NC 4.0 for noncommercial use and helper/server code is Apache-2.0. Separate from SemIf, formerly named OpenJev.

    AnotiaWang/awesome-decision-models

Frequently asked questions

Can I use it commercially?

The weights are CC BY-NC 4.0, so non-commercial only; the code is Apache-2.0.

How many options per question?

Up to 52.

Is it the same as SemIf?

No, it is separate from SemIf, which was formerly named OpenJev.

Related projects

  • Luce

    Open recipe: describe a decision task, have an LLM write training data, and train a Qwen LoRA plus decision head with calibrated outputs.

    ModelbetaPython
  • OpenJev

    OpenJev 1.7B-8B: open decision models distilled for general decisions and browser-agent steps, served behind /v1/systemone.

    ModelbetaPython
  • OpenThai-SystemOne

    Open Thai and English System One model: Qwen3.5-0.8B with a 256-way slot-softmax head, served with a /v1/systemone-compatible API.

    LibrarybetaPython
  • mini-jev

    Dependency-free Python wrapper that turns any LLM API into a Jev-style choose-one-option decision call for old machines.

    ModelprototypePython
  • CUA-S1-FORMS

    706K-parameter Jev-style option scorer that rates fill, check, click or skip for each form element in one parallel pass.

    ModelresearchPython
  • JevForge

    Open stack to synthesise decision data, train and calibrate a Qwen3.5-0.8B scorer, evaluate on Mind2Web and serve Jev-style.

    ServiceresearchPython