OneJev
OneJev: open multimodal System One model (0.8B-27B) answering typed questions about screenshots, photos, video and text in one pass.
Overview
OneJev, from the OmniJev team, is a multimodal System One decision model built on Qwen in 0.8B, 4B, 9B and 27B sizes, with OneJev 1.5 released on 13 October 2026. It returns calibrated probabilities for Choice, Score and Noul questions about screenshots, photos, video and text in a single forward pass, in English, Chinese and Japanese. Weights, a dataset and an online demo are on Hugging Face.
According to the list summary, a question about a 1280x720 screenshot takes 31 ms on the 0.8B and 189 ms on the 27B on one H200, and ten questions cost little more than one. It ships a qev package that serves an API and playground via PyTorch (text, images, video), llama.cpp with GGUF (text and images), or a Docker Compose stack that detects hardware. The example asks questions about a screenshot for a computer-use task like paying an invoice.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which on-screen element or option fits the task?(inferred) | — |
| Yes / no | Is this statement true of the screenshot or video? | yes · no |
| Score | Rate the state on ordered levels | — |
How it works
- Start qev serve with a chosen model (PyTorch, GGUF or Docker).
- Send state, media and typed questions to the local API.
- Read per-option probabilities.
- Use answers to drive computer-use or classification logic.
Strengths
- Multimodal including video (PyTorch backend).
- Several sizes from 0.8B to 27B.
- Multiple questions per pass at little extra cost.
Caveats
- Video requires the PyTorch backend and an NVIDIA GPU.
- Speed figures are from an H200.
Getting started
pip install "qev[torch] @ git+https://github.com/OmniJev/OneJev.git" && qev serve --model OmniJev/OneJev1.5-4B --port 8000At a glance
- Type
- Model
- Status
- beta
- Language
- Python
- Runtime
- PyTorch, llama.cpp, Docker, HTTP server
- Has code
- yes
- License
- Apache-2.0
Use cases and integrations
How curated lists describe it
Open multimodal System One model from the OmniJev team, in four sizes (0.8B to 27B): Choice, Score, and Noul questions about a screenshot, photo, video, or text get a calibrated probability for every option in one forward pass. Weights on Hugging Face.
AnotiaWang/awesome-decision-models🚀🚀 A multimodal System One decision model that gives calibrated answers to typed questions about screens, photos, video and text in one forward pass.
logicrw/awesome-jev-projects🚀🚀 A multimodal System One decision model that gives calibrated answers to…
wh000wh000/awesome-jev-liveOpen multimodal System One model in four sizes (0.8B to 27B): typed questions about a screenshot, photo, video or text get a calibrated probability for every option in one forward pass. On one H200, a question about a 1280x720 screenshot takes 31 ms on the 0.8B and 189 ms on the 27B, and ten questions cost little more than one. On 354 TypeSafe questions with published Jev answers, OneJev-27B scores 89.3 against Jev 1.13's 89.0.
cobanov/awesome-jev🚀🚀 A multimodal System One decision model that gives calibrated answers to typed questions about screens, photos, video and text in one forward pass.
hellogumbo/awesome-jevMultimodal decision model answering typed questions about screenshots, photos, video and text in one forward pass. English, Chinese and Japanese.
MrJev/awesome-jev
Frequently asked questions
What sizes are available?
0.8B, 4B, 9B and 27B.
Can it run on a Mac?
Yes via llama.cpp with GGUF models for text and images; video needs PyTorch.
How fast is it?
Per the list summary, 31 ms for a screenshot question on the 0.8B and 189 ms on the 27B on an H200.
Related projects
Open Jev-style decision models with vision (Qwen3.5 4B/9B LoRA) serving Choice, Score and Noul on /v1/systemone.
LibraryresearchPythonFree hosted Jev-style multimodal decision model API (about 20B active) that picks actions from images and options, aimed at robot control.
ServicebetaComfyUI nodes for Jev-style noul, choice and score decisions over images and text using a local imajev adapter on Qwen3.5.
ProjectbetaPythonResearch code for an image-and-text System One model that reads option-letter logits from a multimodal LLM, with training and calibration.
ProjectresearchPythonIndependent open model on Molmo2-ER that answers spatial questions by choosing from fixed options with one shared classifier head.
ProjectresearchPythonFrom-scratch research reconstruction of a small System-One decision model, used to route and gate commands in a Windows palette.
ProjectresearchPython