The Open Weights
LatestModelsLeaderboardsCompanies
Subscribe
The Open Weights

The daily record of open-source AI. New model releases, leaderboards, and what's coming next — written for people who ship.

Refreshed every 12 hours

Discover

  • Latest releases
  • New today
  • Trending models

Browse

  • All models
  • Companies
  • Categories
  • Leaderboards

About

  • About
  • Editorial policy
  • Hardware estimates
  • RSS feed
  • llms.txt
  • Newsletter

© 2026 The Open Weights. An independent publication.

PrivacyTermsSMSAggregated by Claude · curated by humans.
LatestAutotrustJEV-27B-VL
AutotrustVision-Language

JEV-27B-VL Pairs Vision-Language With Calibrated Odds

A 27B vision-language model from autotrust aims to output typed decisions with probabilities you can actually trust.

Sep 30, 2026
NotableOther
JEV-27B-VL

The team at autotrust has released JEV-27B-VL, a 27-billion-parameter vision-language model built around an unusual goal: producing not just answers, but typed decisions accompanied by calibrated probabilities. The model is now available on Hugging Face.

Most multimodal systems return free-form text, leaving downstream applications to guess how confident the model really is. JEV-27B-VL instead frames its outputs as structured decisions with attached probability estimates — a design that matters most when a wrong answer carries real cost and a system needs to know when to defer, escalate, or abstain.

Why it matters

Calibration is one of the quieter problems in applied AI. A model that is confidently wrong is harder to deploy safely than one that signals uncertainty honestly. By emphasizing calibrated probabilities over raw generations, JEV-27B-VL is positioning itself for workflows where decisions must be auditable.

  • Size: 27B parameters, a dense (non-MoE) architecture
  • Modality: vision-language, handling both images and text
  • Focus: typed decisions with calibrated probability outputs

Key details remain thin — the listing does not specify context length or publish benchmark figures, and the license is marked as "other," so teams evaluating it for production should check the terms on the model card closely. Still, as an opening release for the JEV family, it stakes out a clear and practical niche in the crowded field of open vision-language models.

Sources

  • autotrust/JEV-27B-VL

    Hugging Face

    Visit
OlderAREX-2 arrives as an open deep-research agent modelBAAI · Reasoning · 2 months agoNewerCactus Compute's Whistle brings speech-to-text to the edgeCactus Compute · Speech → Text · 2 months ago

Get the model

Hugging Face

Specs

Parameters27B
Context window262K tokens
Size55.6 GB
PrecisionBF16
ArchitectureQwen3_5ForConditionalGeneration
LicenseOTHER
Downloads893.9K
Likes5.9K

Can you run it?

Runs on a 24 GB GPU at 4-bit.

  • BF16 (as published)

    80 GB GPU (A100 / H100) · 128 GB Mac Studio

    58.7 GB
  • 8-bit

    48 GB GPU (RTX 6000) or 2×24 GB · 64 GB Mac

    32.6 GB
  • 4-bit

    24 GB GPU (RTX 3090 / 4090) · 32 GB Mac

    19.8 GB
Your machine
GGUF builds
  • KV cache sized for 8,192 tokens of context; longer prompts need proportionally more.

Based on Hugging Face weights metadata. Assumes an 8K context and ~1 GB runtime overhead; actual needs vary. How we estimate


Modalities

Vision-Language

The Weekly Weights

Every open release that mattered, one email a week.

0 comments

No comments yet. Be the first to weigh in.

More in Vision-Language

All Vision-Language →
SenseNova-Loop-9B
SenseTime/Vision-LanguageRuns on a laptop

SenseTime Releases Looped 9B Multimodal Model

The company's new SenseNova-Loop-9B pairs vision-language capabilities with a recurrent looped-transformer design.

Nov 17, 2026
Hunyuan Hy4-flash
Tencent/Vision-LanguageMulti-GPU server

Tencent releases Hunyuan Hy4-flash multimodal MoE

The lightweight, mixture-of-experts model adds vision and text instruction-following to Tencent's open Hunyuan lineup.

Nov 17, 2026
Mirage
Cactus Compute/Vision-Language

Cactus Compute's Mirage brings vision tasks on-device

The new open vision-language model targets detection, segmentation, and visual Q&A without relying on the cloud.

Nov 5, 2026