JEV-27B-VL
AutoTrust's vision version of JEV-27B, ranked first of 20 on the community Jev Decision Index vision board.
Overview
JEV-27B-VL is AutoTrust's multimodal decision model: a LoRA adapter on Qwen3.8-27B trained on the jev-distill-corpus-v3 data that answers typed questions about images and text with calibrated probabilities, while keeping a generation path. It is Apache-2.0 licensed and served with vLLM.
The card reports it ranked first of 20 vision decision models on the community Jev Decision Index 0.3 vision board (independent evaluation, updated 7 October 2026), with a full score of 69.82, 72.78 on public benchmarks and 66.86 on private tests built from fresh unpublished images, and an ECE of 0.019. It lists runners-up such as Solomon v1.1 and Xor 1.2 and notes it scores above a stock Qwen3.5-397B VLM reference. A companion blog argues one decision model can replace a ranking team, and a live demo ranks news headlines and re-ranks search results with calibrated decisions in about half a second.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Typed choice about an image and state(inferred) | — |
| Score | How important or relevant is this item? (used for ranking)(inferred) | — |
How it works
- Serve the adapter on Qwen3.8-27B with vLLM
- Send images, state and typed questions
- Receive calibrated probabilities
- Rank, filter or route items by score
Strengths
- Top of an independent vision decision leaderboard per the card
- Low calibration error (ECE 0.019)
- Apache-2.0 open weights
Caveats
- Large model needs significant GPU memory
- Leaderboard positions are snapshots
Getting started
Load autotrust/JEV-27B-VL from Hugging Face with vLLM per the model card.At a glance
- Type
- Model
- Status
- beta
- Language
- Python
- Runtime
- vLLM / transformers
- Has code
- yes
- License
- Apache-2.0
Use cases and integrations
How curated lists describe it
No upstream description was published.
wh000wh000/awesome-jev-live
Frequently asked questions
What does JEV-27B-VL add over JEV-27B?
Image input, so it can make typed decisions about pictures as well as text.
How does it rank against others?
The card reports first of 20 on the Jev Decision Index 0.3 vision board.
Is there a demo?
Yes, a news recommendation and search demo at news.guru.so.
Related projects
Free 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.
ProjectresearchPythonOpen project training a vision-enabled Jev-like model (0.8B, 9B) that judges dynamic candidate sets over images with Choice, Noul and Score.
ProjectresearchPythonIndependent Jev-style server on DiffusionGemma and SGLang that classifies doodles, flowers and messages with Choice, Score and Noul.
ServiceresearchPython