Moment-Video: Benchmark Diagnosing Temporal Fidelity of Video MLLMs on Momentary Visual Events
Moment-Video is a new benchmark of 1,000 human-verified video-QA pairs designed to evaluate how well video multimodal large language models (MLLMs) handle brief, localized visual events that may span only a few frames. The benchmark covers 7 domains and 25 subcategories across four task types: Temporal Occurrence, Temporal Counting, Action Description, and Temporal Reasoning. Evaluation of 33 proprietary and open-source models reveals severe deficiencies: the best model (Seed-2.0-Pro) achieves only 39.6% accuracy, while most open-source models score below 25%. Diagnostic analyses show that denser frame sampling helps but does not resolve the bottleneck, pointing to fundamental limitations in how current video MLLMs represent and preserve transient visual evidence.
Related guides (3)
Related events (8)
TempGlitch: Benchmark for Evaluating VLMs on Temporal Glitch Detection in Gameplay Videos
TempGlitch is a new benchmark designed to evaluate vision-language models on temporal glitch detection in gameplay videos, distinguishing temporal anomalies (visible only across ordered frames) from spatial ones (visible in a single frame). The benchmark covers five temporal glitch types with paired glitch-free videos for binary evaluation, and tests 12 proprietary and open-weight VLMs across multiple frame-sampling settings. Results show current VLMs perform near chance on temporal glitches, with neither denser frame sampling nor larger model size reliably improving detection. The work highlights a systematic gap in VLM temporal reasoning capabilities relevant to automated video quality assurance.
TimeScope: How Long Can Your Video Large Multimodal Model Go?
Hugging Face introduces TimeScope, a benchmark designed to evaluate video large multimodal models (LMMs) across varying video lengths and temporal reasoning demands. The benchmark targets a known gap in existing evaluations: most video benchmarks use short clips, leaving long-video understanding largely untested. TimeScope aims to systematically probe how model performance degrades or holds as video duration increases.
PaSBench-Video: A Streaming Video Benchmark for Proactive Safety Warning in MLLMs
PaSBench-Video is a 740-video benchmark designed to evaluate whether multimodal large language models can issue timely, accurate safety warnings during the window between a visible danger sign and an accident. Videos span four domains (driving, healthcare, daily life, industrial production) and are annotated with frame-level risk onset and accident boundaries, requiring causal temporal reasoning rather than static scene classification. Testing 13 MLLMs reveals no model exceeds 20% on the strictest metric, with recall strongly coupled to false-positive rate (Pearson r=0.64), indicating models rely on scene-level activity cues rather than genuine hazard reasoning. Performance varies sharply by domain, with driving being particularly problematic due to visual similarity between routine and hazardous scenes.
VideoMLA: Low-Rank Latent KV Cache for Minute-Scale Autoregressive Video Diffusion
VideoMLA applies Multi-Head Latent Attention (MLA) to causal video diffusion, replacing per-head keys and values with a shared low-rank content latent and decoupled 3D-RoPE positional key, achieving 92.7% reduction in per-token KV memory. The paper investigates why MLA works despite pretrained video attention not being low-rank (unlike the spectral assumption motivating MLA in LLMs), finding that the MLA bottleneck itself determines effective rank rather than the pretrained spectrum. On VBench, VideoMLA matches short-horizon baselines, achieves best overall score at long horizons, and delivers 1.23x throughput improvement on a single NVIDIA B200 GPU.
Survey: Human-View Video Understanding with MLLMs — Watch, Remember, Reason Framework
A new arXiv survey paper proposes a unified 'human-view' framework for analyzing multimodal LLM-based video understanding, organized around three functional abilities: watching (perception), remembering (memory), and reasoning. The authors introduce a formulation characterizing video understanding systems by perceptual representations, memory states, reasoning traces, and predictions, then survey methods, datasets, and benchmarks across these dimensions. The work covers challenges including spatio-temporal perception, long-video processing, streaming understanding, and faithful reasoning, with application domains spanning egocentric, sports, medical, and narrative video.
AdaCodec: Predictive Visual Coding for Efficient Video MLLMs
AdaCodec introduces a predictive visual code interface for video multimodal large language models that exploits temporal redundancy in video. Instead of encoding every sampled frame as an independent RGB image, it sends full visual tokens only for reference frames with high conditional predictive cost, and encodes inter-frame changes as compact P-tokens. Evaluated against a Qwen3-VL-8B per-frame baseline across eleven benchmarks, AdaCodec at 1/7 the token budget (32k vs 224k tokens) surpasses the baseline on all long-video benchmarks while reducing time-to-first-token from 9.26s to 1.62s.
Benchmark for view-level visual evidence identification in multi-view MLLMs for autonomous driving
A new arXiv preprint introduces a multi-view visual question answering benchmark targeting evidence-source identification in autonomous driving scenarios. Given six synchronized NuScenes camera views and a question, models must identify which camera view supports the answer — not just produce a correct answer. The 122-pair benchmark spans causality, counterfactual reasoning, and intent prediction, and exposes grounding failures that answer-only evaluation misses. The work addresses a meaningful gap between answer accuracy and correct visual grounding in safety-critical multimodal systems.
STORM: Internalized Spatial-Temporal Reasoning for Video-Language Models via Latent Trajectories
STORMS is a two-stage training framework that teaches large vision-language models to perform spatial-temporal video reasoning through bounded continuous latent trajectories rather than explicit textual chain-of-thought, keyframe selection, or external tool use. In Stage I, latent tokens are aligned with thought-video representations derived from generated videos; in Stage II, answer-only supervision internalizes the reasoning process. At inference time, no video regeneration or frame reinsertion is required, reducing latency and engineering complexity. Evaluations on VideoMME, MVBench, TempCompass, and MMVU show improved accuracy with substantially lower inference overhead versus tool-based pipelines.