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

Open-Jev

Open-Jev: open typed-decision models (2B, 9B, 27B v1.1) returning candidate probabilities without generation, with a no-code workbench.

Modelbeta
View on GitHub ↗Live demo ↗

Overview

Open-Jev provides open typed-decision models: supply a context, questions and candidates and get typed probabilities directly, without autoregressive generation or parsing JSON. The current flagship is Open-Jev-27B-v1.1 on Qwen3.8-27B, alongside 2B and 9B models and public datasets on Hugging Face. A project site hosts demos, measured results and background.

The v1.1 demo replays real model outputs on synthetic held-out confirmation-gate cases, showing how seven possible decisions are ranked when a misleading preview changes; the README notes it is edited playback and not a controlled experiment. A no-code workbench lets you paste messages or import a CSV, define categories and download suggested labels with probabilities, running the 2B model on CPU in a Hugging Face Space. Other projects report Open-Jev-27B-v1.1 at 197/231 on public JevBench. It is for teams wanting open, self-hostable decision models.

What the decision model decides

TypeQuestionAnswers
ChoiceWhich of the candidate decisions fits this context (e.g. confirmation gate)?—
ChoiceWhich category does this CSV row belong to?—

How it works

  1. Choose a model size from Hugging Face.
  2. Send context, questions and candidates to the local server.
  3. Receive per-candidate probabilities.
  4. Or use the workbench to label a CSV without code.

Strengths

  • Open weights and datasets in several sizes.
  • No-code CSV workbench.
  • Transparent demo evidence.

Caveats

  • 27B model needs substantial GPU resources.
  • Hosted workbench on CPU is slow.

Getting started

Try the workbench at https://zefan-cai.github.io/open-jev/workbench/ or download ZefanCai/Open-Jev-27B-v1.1 and run python -m jev.server.

At a glance

Type
Model
Status
beta
Language
Python
Runtime
local server, web workbench
Has code
yes

Use cases and integrations

How curated lists describe it

  • Open typed-decision models that return probabilities over candidates directly, without generating or parsing an answer, with a project site and reports.

    MrJev/awesome-jev

Frequently asked questions

What sizes are available?

2B, 9B and 27B (v1.1).

Can I use it without code?

Yes, the workbench labels pasted messages or CSV rows.

Is the demo a live run?

No, it is edited playback of real saved model outputs.

Related projects

  • brier

    Mangaba AI's local, calibrated typed-decision model for Portuguese text, trained to optimize the Brier score; proprietary since v5.

    ProjectproductionPython
  • blink★ 22

    Open one-pass typed-decision model with a dependency-free C99 runtime that also builds to a 66 KB WebAssembly module.

    ServicebetaC
  • Foq★ 11

    Local open alternative to Jev using a 1.58-bit ternary 8B model to answer booleans, choices, scores and Pydantic extraction in ~25 ms.

    ProjectbetaPython
  • browser-laya

    Runs the open Laya decision model entirely in a browser tab with WebGPU and ONNX Runtime Web: no server, no API key.

    ServicebetaTypeScript
  • ComfyUI-TypedDecision

    ComfyUI nodes for Jev-style noul, choice and score decisions over images and text using a local imajev adapter on Qwen3.5.

    ProjectbetaPython
  • ifReflex.cpp

    llama.cpp-based C++ server that turns any LLM into a System One decision model, reading answers from next-token logits.

    ProjectbetaC++