Gemma-4-26B-A4B-it (AI model): news timeline & CVEs

autotrust/GEV-26B-Decide — new model trending #30 on Hugging Face

AutoTrust released GEV-26B-Decide, an open 26B Gemma-4 decision model scoring 62.48 on Decision Index 0.2.1.

AutoTrust AI published autotrust/GEV-26B-Decide, open weights previously named JEV-Gemma4-26B-A4B, built on Google's Gemma-4-26B-A4B-it (26 billion parameters, about 4 billion active). On Decision Index 0.2.1 it scores 62.48 balanced skill, above the TypeSafe Jev 1.13 board mark of 57.91; area scores include Tools and Automation 0.697 and Arts and Taste 0.415. System 1 returns calibrated option probabilities in one pass, about 45 ms on a B200, and adaptive thinking invokes Gemma-4 reasoning below 0.8 confidence. In demos it finished 95% of 60 browser tasks at about 85 ms per click, but only 40% of 20 robot-arm pick-and-place scenes.

Hugging Face trending modelsupdated Oct 5, 01:31 UTC · 4d agofirst Oct 2, 04:20 UTC · 6d agoModel release 2 sources1

From Routing Signals to Selective Review: Visual regrounding in MoE VLMs

MoE routing signals detect missing visual targets in Qwen and Gemma VLMs and trigger selective review.

The paper uses Mixture-of-Experts routing probabilities to detect target-absence grounding failures in vision-language models before generation. Linear detectors on Qwen3-VL-30B-A3B-Instruct and Gemma-4-26B-A4B-it reach ROC-AUCs of 0.9988 and 0.9956 on GQA-Inpaint, retaining 0.8095 and 0.7781 on OBER. A routing-gated review prompt raises end-to-end accuracy by up to 22.25% for Qwen and 13.42% for Gemma without modifying weights.

arXiv cs.AI / cs.LG / cs.CLSep 29, 17:49 UTC · 9d agoAI research

Gemma-4-26B-A4B-it

Timeline2 articles

Appears with

autotrust/GEV-26B-Decide — new model trending #30 on Hugging Face

From Routing Signals to Selective Review: Visual regrounding in MoE VLMs