Video-MemComp - VideoMME Overall: 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. VideoMME Overall. Metric: Accuracy (%).

Metric: Accuracy (%). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 21 models tracked.

Top models

#ModelScoreOverall rank
1Video-MemComp GPT4-o [frames 1fps; KV –]77.2
2Video-MemComp Qwen2-VL-7B [frames 768; KV 100K]63.7
3Video-MemComp Qwen2-VL-7B + InfiniPot-V [frames 768; KV 6K]62.8
4Video-MemComp Qwen2-VL-7B + StreamMem [frames 0.5/0.2fps; KV 6K]62.1
5Video-MemComp Qwen2.5-VL-3B + Video-MemComp [frames 768; KV 6K]61.2
6Video-MemComp Qwen2-VL-7B + Video-MemComp [frames 768; KV 6K]60.8
7Video-MemComp GPT4-V [frames 1fps; KV –]60.7
8Video-MemComp LongVU 7B [frames 1fps; KV 8K]60.6
9Video-MemComp Qwen2-VL-7B + DyCoke [frames 768; KV 6K]59.8
10Video-MemComp Qwen2.5-VL-3B + StreamMem [frames 4.0/0.5 fps; KV 6K]59.5

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