Video-MemComp Intra-frame Ablation - MLVU: leaderboard
Video-MemComp v1: streaming evaluation of official development sets. Table 1 mixes reproduced starred rows and results from prior papers; frame sampling and average KV budget differ by system and remain in row identity. Backbone checkpoint revision and decoding are not fully specified; source-qualified systems are retained. This is video QA under cache compression, not an unmodified backbone score. Section 4.4 explicitly binds these experiments to Qwen2.5-VL-3B, VideoMME streaming at 4.0/0.5 fps and MLVU full 768 frames. Delta row is not a model. MLVU. Metric: Accuracy (%).
Metric: Accuracy (%). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 2 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Video-MemComp Qwen2.5-VL-3B inter plus intra-frame | 68.19 | |
| 2 | Video-MemComp Qwen2.5-VL-3B inter-frame only | 67.68 |
Interactive version: theaggregate.ai/benchmark?slug=video-memcomp-intra-frame-ablation-mlvu · How It Works · Data refreshed daily, snapshot 2026-10-27.