Leum-VL Technical Report

Yuxuan He; Chaiming Huang; Yifan Wu; Hongjun Wang; Chenkui Shen; Jifan Zhang; Long Li

cs.MMMar 20, 2026

Leum-VL Technical Report

Authors: Yuxuan He, Chaiming Huang, Yifan Wu, Hongjun Wang, Chenkui Shen, Jifan Zhang, Long Li

Organizations: Hainan Sihe Data Technology Co., Ltd.

Abstract

A short video succeeds not simply because of what it shows, but because of how it schedules attention -- yet current multimodal models lack the structural grammar to parse or produce this organization. Existing models can describe scenes, answer event-centric questions, and read on-screen text, but they are far less reliable at identifying timeline-grounded units such as hooks, cut rationales, shot-induced tension, and platform-facing packaging cues. We propose SV6D (Structured Video in Six Dimensions), inspired by professional storyboard practice in film and television production, a representation framework that decomposes internet-native video into six complementary structural dimensions -- subject, aesthetics, camera language, editing, narrative, and dissemination -- with each label tied to physically observable evidence on the timeline. We formalize a unified optimization objective over SV6D that combines Hungarian-matched temporal alignment, dimension-wise semantic label distance, and quality regularization. Building on this framework, we present Leum-VL-8B, an 8B video-language model that realizes the SV6D objective through an expert-driven post-training pipeline, further refined through verifiable reinforcement learning on perception-oriented tasks. Leum-VL-8B achieves 70.8 on VideoMME (w/o subtitles), 70.0 on MVBench, and 61.6 on MotionBench, while remaining competitive on general multimodal evaluations such as MMBench-EN. We also construct FeedBench, a benchmark for structure-sensitive short-video understanding. Our results indicate that the missing layer in video AI is not pixel generation but structural representation: grounded on the timeline, linked to visible evidence, and directly consumable by downstream workflows such as editing, retrieval, recommendation, and generation control, including text-heavy internet video formats with overlays and image-text layouts.

Figures & tables

Figure 1: Video understanding benchmark comparison of Leum-VL-8B against 8B-scale models.

Figure 2: Overview of the data synthesis pipeline. Starting from raw internet video, the pipeline proceeds through shot segmentation, LoRA-based structural annotation, scorer-based quality filtering, LLM-assisted structural extension, and dense report generation, producing the SV6D SFT training corpus.

Dimension	Sub-dimensions
Camera language	Shot size
	Camera position
	Shooting angle
	Lens focal length
	Camera movement
	Depth of field

Table 1: Per-dimension label taxonomy of FeedBench. Sub-dimensions are listed for each of the six structural dimensions.

Dimension	Temporal anchor	Task Types	# Labels	# Items
Subject	Interval	Single, Ordered	8	88 †
Aesthetics	Interval	Single, Multi, Ordered	41	6,645
Camera language	Interval	Single, Multi, Ordered	38	5,927
Editing	Interval / Timestamp	Single, Multi, Ordered	65	949
Narrative	Interval	Single, Multi, Ordered	[WIP]	[WIP]
Dissemination	Interval	Single, Multi	40	194

Table 2: Overview of FeedBench, containing 60 short-form videos from 27 content categories. All core items are grounded in either a temporal interval or a timestamp, instantiated as closed-vocabulary multiple-choice or IoU-scored temporal localization tasks. † Subject items are currently IoU-only (temporal localization); narrative items are under construction ([WIP]). Single, Multi, and Ordered denote single-label, multi-label, and ordered-transition items, respectively.

Parameter	Value
Group size $G$	8
Prompts per step	390 (13 task types $\times$ 30)
Sequences per step (GBS)	3,120 ( $390\times 8$ )
PPO mini-batch size	390 (single gradient step)
Learning rate	$1\times 10^{-6}$ (cosine $\to 1\times 10^{-7}$ )
KL penalty	disabled

Table 3: GRPO training hyperparameters.

Task Type	Reward
Temporal grounding	$0.2\,\mathrm{Format}+0.4\,\mathrm{IoU}+0.4\,\mathrm{Label}$
Temporal action localization	$0.2\,\mathrm{Format}+0.8\,\mathrm{IoU}$
OCR (regular & handwritten)	$\mathrm{EditDist}(\hat{y},\,y^{*})$
Chain-of-thought	$0.3\,\mathrm{Format}+0.7\,\mathrm{Judge}$

Table 4: Reward function by task type.

Figure 3: GRPO training dynamics over 150 steps. Left : actor entropy decreases steadily, indicating the policy becomes more confident. Center : mean reward score rises from

\sim

0.66 to

\sim

0.80. Right : mean response length remains stable, showing no reward hacking through verbosity.

Model	Subject	Aesthetics	Camera Language	Editing	Narrative	Dissemination
Qwen3-VL-8B-Instruct †	62.3	44.0	42.0	49.2	–	44.9
Qwen3.5-9B ( Team, 2025b ) †	69.6	52.3	42.0	56.4	–	56.4
Leum-VL-8B	78.3	64.8	53.0	74.8	–	71.1

Table 5: Performance of Leum-VL-8B on FeedBench.

Category	Benchmark	Leum-VL-8B	Qwen3-VL-8B 1	Keye-VL-8B Thinking 2	GLM-4.1V-9B Thinking 3	MiniCPM-V-4.5-8B 4
General VQA	MMBench-EN ${}_{\text{test}}$	84.8	84.5	92.0	85.8	84.2
	MMBench-CN ${}_{\text{test}}$	83.9	84.7	92.0	84.7	84.2
	HallusionBench	56.5	61.1	62.7	63.2	61.2
	RealWorldQA	73.2	71.5	73.5	—	72.1
	MMStar	67.5	70.9	80.5	72.9	72.1
	BLINK	65.2	69.1	54.9 ‡	65.1	42.0 ‡

Table 6: Performance of Leum-VL-8B and other 8B-scale models on open benchmarks.

Video Understanding

Video understanding develops methods for recognizing actions, tracking events, grounding evidence in time, and reasoning across visual and audio streams.

See all 116 papers

HAVEN: Hierarchically Aligned Multimodal Benchmark for Unified Video Understanding
May 19, 2026Mengqi Shi, Haopeng ZhangMultimodal Grounding Video Summarization
Department of Information and Computer Sciences · University of Hawaii at Manoa
Improving Spatial-Temporal Reasoning in Video-Language Models with Structured Video Prompting
Aug 23, 2026Sadegh MohammadianVideo-Language Models Structured Prompting
Sharif University of Technology
Moment-Video: Diagnosing Temporal Fidelity of Video MLLMs on Momentary Visual Events
Jun 1, 2026Xiaolin Liu, Yilun Zhu, Xiangyu Zhao +9VLM Evaluation Video QA
Shandong University · Southeast University · Shanghai Jiao Tong University +1

May 19, 2026J/K move · Enter open · S save

HAVEN: Hierarchically Aligned Multimodal Benchmark for Unified Video Understanding

Mengqi Shi, Haopeng Zhang

Department of Information and Computer Sciences · University of Hawaii at Manoa

Multimodal Grounding Video Summarization Multimodal Model Evaluation Temporal Video Understanding Multimodal Understanding

Open paper AskPDF arXiv

While Multimodal Large Language Models (MLLMs) exhibit strong performance on standard video tasks, their ability to faithfully summarize and reason over complex narratives remains poorly evaluated. Existing summarization benchmarks fragment supervision across isolated granularities, such as keyframes, key shots, or disjointed text summaries, failing to capture the inherently hierarchical structure of cross-modal alignment. To address this critical gap, we introduce HAVEN, a hierarchically aligned multimodal benchmark for unified video understanding. HAVEN pioneers a fully granular (frame, shot, and video levels) and fully multimodal (video and text) dataset architecture, complete with explicit, continuous alignment between modalities. Built upon this unified annotation paradigm, we propose a comprehensive evaluation suite spanning summarization, temporal reasoning, multimodal grounding, and saliency ranking. Extensive benchmarking of state-of-the-art MLLMs exposes a persistent gap between surface-level textual fluency and grounded multimodal understanding. Ultimately, HAVEN advances the evaluation of multimodal systems beyond traditional QA formats, offering a rigorous, standardized testbed to drive future research in interpretable, hierarchical video understanding. We publicly release the dataset, benchmark suite, and evaluation protocols.

Abstract

Video Understanding

Explore similar work

HAVEN: Hierarchically Aligned Multimodal Benchmark for Unified Video Understanding

Improving Spatial-Temporal Reasoning in Video-Language Models with Structured Video Prompting

Moment-Video: Diagnosing Temporal Fidelity of Video MLLMs on Momentary Visual Events