Video-MemComp - 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. MLVU. Metric: Accuracy (%).
Metric: Accuracy (%). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 21 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Video-MemComp LLaVA-OneVision-7B + ReKV [frames 0.5 fps; KV 6K] | 68.5 | |
| 2 | Video-MemComp Qwen2.5-VL-3B + Video-MemComp [frames 768; KV 6K] | 68.2 | |
| 3 | Video-MemComp Qwen2-VL-7B + Video-MemComp [frames 768; KV 6K] | 67.1 | |
| 4 | Video-MemComp LLaVA-OneVision-7B + StreamMem [frames 0.5/0.2 fps; KV 6K] | 66.9 | |
| 5 | Video-MemComp LLaVA-OneVision-7B + Video-MemComp [frames 0.5/0.2 fps; KV 6K] | 66.5 | |
| 6 | Video-MemComp LLaVA-OneVision-7B + LiveVLM [frames 0.5/0.2 fps; KV -] | 66.3 | |
| 7 | Video-MemComp GPT4-o [frames 1fps; KV –] | 66.2 | |
| 8 | Video-MemComp Qwen2-VL-7B + StreamMem [frames 0.5/0.2fps; KV 6K] | 65.9 | |
| 9 | Video-MemComp Qwen2-VL-7B [frames 768; KV 100K] | 65.8 | |
| 10 | Video-MemComp Qwen2-VL-7B + InfiniPot-V [frames 768; KV 6K] | 65.8 |
Interactive version: theaggregate.ai/benchmark?slug=video-memcomp-mlvu · How It Works · Data refreshed daily, snapshot 2026-10-27.