Skip to content
Trending storyDeveloping

VideoJev-4B video decision model released under Apache 2.0 license

1 article1 sourcesince Oct 19Last article 58m ago ·

Overview

AISummary of 1 article

ModelScope releases VideoJev-4B, a 4B-parameter video decision model based on Qwen3.5-4B from MCG-NJU, for video question answering and temporal grounding.

The model scores candidate answers and time spans directly. ModelScope says it ranks first among sub-6B models on LVBench, Video-MME, LongVideoBench, and MLVU, and supports up to 256K context in the official video QA evaluation setup.

The release includes BF16 weights, training and evaluation code, and a quick-start demo, under the Apache 2.0 license.

Written by AI from the articles below · updated Oct 19, 10:30 PM ET

Check the sources:

Latest developmentOct 19, 10:18 PM ET
VideoJev-4B, a 4B video decision model for QA and temporal grounding

Article timeline

The articles in this story. Times are ET.

Oct 19
  1. ModelScopeOfficial
    VideoJev-4B, a 4B video decision model for QA and temporal grounding

    AIVideoJev-4B, a Qwen3.5-4B-based model from MCG-NJU, scores candidate answers and time spans directly for video question answering and temporal grounding. It ranks first among sub-6B models on LVBench, Video-MME, LongVideoBench, and MLVU, and supports up to 256K context in the official video QA evaluation setup. The model is released under Apache 2.0 with BF16 weights, training and evaluation code, and a quick-start demo.

    Image from @ModelScope2022's post

Heat Trend

Not enough continuous observations to show a trend yet.