JEV-27B-VL — vision variant of the 27B decision model
A vision-language variant of the JEV 27B System-1 decision model: typed decisions (yes/no, pick-one, rate) over text and images in one forward pass, ranked #1 of 20 vision decision models on the Jev Decision Index Vision board.
Why it matters
A decision head trained on text only, zero-shot on images, beating a 397B reference VLM (+6.4 points) with calibrated probabilities (ECE 0.019) — plus working robot-arm and computer-use demos where every step is one visual decision.
What you could build with it
A robotics team could use the model's single-forward-pass visual decisions as a fast policy gate — e.g. 'is the target left or right of the gripper' from a camera frame at ~240 ms per decision — before handing execution to a heavier controller.
Does it hold up?
Too early to judge beyond benchmarks — the model card documents robot-arm (75% of 20 scenes) and computer-use (95% of 60 tasks) demos with published per-episode results, but these are author-run; no independent replications yet.
Built with JEV-27B-VL
- Jev Decision Index Vision board (updated Oct 7, 2026)huggingface · Independent community evaluation ranks JEV-27B-VL first of 20 vision decision models on chance-corrected skill across 20 benchmark columns, including private tests built from unpublished images.
- autotrust/JEV-27B-VL: a decision model that learned to see without a single image of training (HF blog)huggingface · Release writeup (by the model authors) documenting real built demos: zero-shot image recommendation matching 59k-user collaborative filtering, multimodal judge demos with code, VL-RewardBench results.
Learn more
First spotted on hf: source.
More AI models
OpenAI publishes 722 AI-generated math manuscripts from an unreleased modelOpenAI released 722 mathematical manuscripts in 372 result families to GitHub under Apache-2.0, produced by an…models · JEV 0.74Mistral Large 4 (Le Chonk): 1.05T-param open-weight MoEMistral's new flagship: a 1.05T-parameter mixture-of-experts multimodal model with 49B active params and 1M-token…models · JEV 0.72ElevenLabs Eleven v4 + v4 Turbo: new emotive TTS architecture with a 100ms real-time variantElevenLabs released Eleven v4 and Eleven v4 Turbo on September 28: a new text-to-speech architecture with inline…models · JEV 0.72VoiceStudio: open-source, fully-local ElevenLabs alternative with voice cloning, dubbing and dictation in 646 languagesVoiceStudio is an open-source, fully-local voice AI studio: voice cloning from a 3-second sample, voice design, video…models · JEV 0.71Mistral Forge: enterprise platform for training and continuously improving proprietary modelsMistral AI launched Forge (Sep 28) — a full-lifecycle model training platform (pre-training, SFT, DPO/ODPO, RL…models · JEV 0.69DeepSeek V4-Flash official API: public beta with upgraded agent capabilities, Responses API and Codex supportDeepSeek's official V4-Flash is now a public beta API with massively upgraded agent capabilities - benchmark scores…models · JEV 0.69
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.