aiqbench analysis
A new benchmark tests whether video multimodal LLMs can detect and reason about brief, answer-critical events, a capability broader video-understanding scores don't measure. The best of 34 evaluated models reaches 39.6% accuracy, and most open-source models score below 25%. The authors also find that denser frame sampling doesn't fix the errors and that performance drops on longer videos, which points to temporal localization as a real gap for anyone building or deploying video AI.
Moment-Video: Diagnosing Temporal Fidelity of Video MLLMs on Momentary Visual Events
arXiv:2606.02522v2 Announce Type: replace-cross Abstract: Video multimodal large language models (MLLMs) have made rapid progress in general and long-form video understanding, yet their ability to capture and use brief, answer-critical visual evidence remains underexplored. Momentary actions or state transitions may determine an answer while occupying only a small fraction of the video. Performance on broad video understanding tasks alone does not establish whether models can reliably identify and use this temporally localized evidence. We introduce Moment-Video, a benchmark for evaluating temporal fidelity through momentary visual event understanding. Questions are constructed around visually observable events or event sequences, requiring models to notice, count, describe, or reason about transient evidence. The benchmark contains 1,000 human-verified video-QA pairs across seven domains and 25 subcategories, covering four tasks: Temporal Occurrence, Temporal Counting, Action Description, and Temporal Reasoning. We evaluate 34 proprietary and open-source MLLMs. Under the main evaluation settings, the best-performing model, Seed-2.0-Pro, achieves 39.6% overall accuracy, while most open-source models remain below 25%. Diagnostic analyses show that denser frame sampling improves some models, but substantial errors persist even when sampled frames cover all annotated necessary event intervals in an audited subset. Performance also declines on longer videos, suggesting additional temporal-localization challenges. By exposing these limitations, Moment-Video provides a diagnostic benchmark to guide the evaluation and development of video MLLMs that more reliably understand brief, answer-critical visual events.
Abstract:Video multimodal large language models (MLLMs) have made rapid progress in general and long-form video understanding, yet their ability to capture and use brief, answer-critical visual evidence remains underexplored. Momentary actions or state transitions may determine an answer while occupying only a small fraction of the video. Performance on broad video understanding tasks alone does not establish whether models can reliably identify and use this temporally localized evidence. We introduce Moment-Video, a benchmark for evaluating temporal fidelity through momentary visual event understanding. Questions are constructed around visually observable events or event sequences, requiring models to notice, count, describe, or reason about transient evidence. The benchmark contains 1,000 human-verified video-QA pairs across seven domains and 25 subcategories, covering four tasks: Temporal Occurrence, Temporal Counting, Action Description, and Temporal Reasoning. We evaluate 34 proprietary and open-source MLLMs. Under the main evaluation settings, the best-performing model, Seed-2.0-Pro, achieves 39.6% overall accuracy, while most open-source models remain below 25%. Diagnostic analyses show that denser frame sampling improves some models, but substantial errors persist even when sampled frames cover all annotated necessary event intervals in an audited subset. Performance also declines on longer videos, suggesting additional temporal-localization challenges. By exposing these limitations, Moment-Video provides a diagnostic benchmark to guide the evaluation and development of video MLLMs that more reliably understand brief, answer-critical visual events.
Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
arXiv:2606.02522 [cs.CV]
(or arXiv:2606.02522v2 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2606.02522 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 317 words
Related coverage
NarrativeTrack: Evaluating Entity-Centric Reasoning for Narrative Understanding
arXiv · Aug 4, 2026
LVSum: A Benchmark for Timestamp-Aware Long Video Summarization
Apple · Jul 20, 2026
LVSum: A Benchmark for Timestamp-Aware Long Video Summarization
arXiv · Jul 20, 2026
STRAND: Benchmarking and Improving Object-Centric Spatio-Temporal Monitoring in Video Large Language Models
arXiv · Sep 25, 2026