Video-MemComp - VideoMME Long KV 12000: 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.2 specifies Qwen2.5-VL-3B on long VideoMME; exact KV budget 12000. Metric: Accuracy (%).
Metric: Accuracy (%). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Video-MemComp Qwen2.5-VL-3B Video-MemComp Two-Stage | 53.44 | |
| 2 | Video-MemComp Qwen2.5-VL-3B Token Compression | 52.89 | |
| 3 | Video-MemComp Qwen2.5-VL-3B KV Cache Compression | 51.11 | |
| 4 | Video-MemComp Qwen2.5-VL-3B StreamMem | 50 | |
| 5 | Video-MemComp Qwen2.5-VL-3B InfiniPot-V | 49.67 |
Interactive version: theaggregate.ai/benchmark?slug=video-memcomp-videomme-long-kv-12000 · How It Works · Data refreshed daily, snapshot 2026-10-27.