Video-MemComp OVO - HLD: 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: choice-based streaming OVO-Bench; Qwen2.5-VL-7B compression hyperparameters tuned for this benchmark, 4K KV versus approximately50K baseline. HLD. Metric: Accuracy (%).
Metric: Accuracy (%). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Video-MemComp Qwen2.5-VL-7B Video-MemComp [OVO; frames 1fps; KV 4K] | 36.6 | |
| 2 | Video-MemComp Qwen2-VL-7B [OVO; frames 64; KV ≈ \approx 10K] | 34.4 | |
| 3 | Video-MemComp InternVL-V2-8B [OVO; frames 64; KV ≈ \approx 10K] | 27.4 | |
| 4 | Video-MemComp Qwen2.5-VL-7B [OVO; frames 1fps; KV ≈ \approx 50K] | 26.9 | |
| 5 | Video-MemComp LLaVA-OneVision-7B [OVO; frames 64; KV ≈ \approx 10K] | 23.7 | |
| 6 | Video-MemComp LLaVA-NeXT-7B [OVO; frames 64; KV ≈ \approx 10K] | 9.7 | |
| 7 | Video-MemComp TimeChat-Online-7B [OVO; frames 1fps; KV ≈ \approx 5K] | 9.1 |
Interactive version: theaggregate.ai/benchmark?slug=video-memcomp-ovo-hld · How It Works · Data refreshed daily, snapshot 2026-10-27.