Video-MemComp - MLVU KV 24K: 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. Appendix E.1: Qwen2-VL-7B, MLVU dev, matched global KV capacity equivalent to 24K per layer. Full-KV row is a single reference spanning columns and is not replicated into budgets. MLVU KV 24K. Metric: Accuracy (%).
Metric: Accuracy (%). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Video-MemComp Qwen2-VL-7B Video-MemComp | 69.43 | |
| 2 | Video-MemComp Qwen2-VL-7B StreamMem | 66.3 | |
| 3 | Video-MemComp Qwen2-VL-7B InfiniPot-V | 65.73 | |
| 4 | Video-MemComp Qwen2-VL-7B SnapKV | 64.37 | |
| 5 | Video-MemComp Qwen2-VL-7B Uniform | 64.18 | |
| 6 | Video-MemComp Qwen2-VL-7B InfiniPot | 63.71 |
Interactive version: theaggregate.ai/benchmark?slug=video-memcomp-mlvu-kv-24k · How It Works · Data refreshed daily, snapshot 2026-10-27.