Video-MemComp OVO Compression Thresholds - REC: 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 F identifies Qwen2.5-VL-7B on OVO. These are distinct hyperparameter arms evaluated on the same questions, not independently sampled tasks; preserve every configuration. REC. Metric: Accuracy (%).

Metric: Accuracy (%). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 16 models tracked.

Top models

#ModelScoreOverall rank
1Video-MemComp Qwen2.5-VL-7B Compression Thresholds [0.30, 0.75]36.73
2Video-MemComp Qwen2.5-VL-7B Compression Thresholds [0.80, 0.85]36.5
3Video-MemComp Qwen2.5-VL-7B Compression Thresholds [0.95, 0.90]36.43
4Video-MemComp Qwen2.5-VL-7B Compression Thresholds [0.70, 0.85]36.1
5Video-MemComp Qwen2.5-VL-7B Compression Thresholds [0.60, 0.85]35.66
6Video-MemComp Qwen2.5-VL-7B Compression Thresholds [0.75, 0.80]35.65
7Video-MemComp Qwen2.5-VL-7B Compression Thresholds [0.90, 0.80]35.25
8Video-MemComp Qwen2.5-VL-7B Compression Thresholds [0.50, 0.95]35.23
9Video-MemComp Qwen2.5-VL-7B Compression Thresholds [0.70, 0.80]34.8
10Video-MemComp Qwen2.5-VL-7B Compression Thresholds [0.80, 0.75]34.52

Interactive version: theaggregate.ai/benchmark?slug=video-memcomp-ovo-compression-thresholds-rec · How It Works · Data refreshed daily, snapshot 2026-10-27.