Video-HolmesV2: leaderboard
Long-video reasoning with spatio-temporal audio-visual evidence: the overall score across perception and reasoning modules, weighted by each module's sample size. Higher is better. Source: arxiv.org/abs/2609.17248.
Metric: Overall (%). Source: arxiv.org. Saturation forecast: Around February 2027. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 73.2 | #159 |
| 2 | Gemini 2.5 Flash | 65.3 | #255 |
| 3 | Qwen 3 VL 235B A22B | 55.3 | #251 |
| 4 | InternVL3-8B | 48.1 | #624 |
| 5 | GPT-4o | 46.9 | #357 |
| 6 | Qwen2.5-Omni-7B | 40.7 | #633 |
| 7 | Qwen 2.5 VL 7B | 38.1 | #641 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=video-holmesv2 · How It Works · Data refreshed daily, snapshot 2026-11-05.