Video-Holmes: leaderboard
Complex video reasoning benchmark over suspense short films, requiring models to actively find and connect clues across video segments.
Metric: Overall (imported). Source: huggingface.co. Source: benchmarklist.com. Saturation forecast: Estimated already saturated. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 45 | #152 |
| 2 | Gemini 2.0 Flash (Thinking) | 43.1 | #342 (Gemini 2.0 Flash) |
| 3 | GPT-4o | 42 | #343 |
| 4 | Gemini 1.5 Pro | 41.2 | #445 |
| 5 | Claude 3.7 Sonnet | 41 | #249 |
| 6 | Claude 3.5 Sonnet | 39.3 | #350 |
| 7 | InternVL3-8B | 32.3 | #620 |
| 8 | Gemini 2.0 Flash | 30.6 | #342 |
| 9 | O4 Mini | 29.9 | #180 |
| 10 | Qwen 2.5 VL 7B | 27.8 | #643 |
| 11 | Qwen2.5-Omni-7B | 16.4 | #624 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=video-holmes · How It Works · Data refreshed daily, snapshot 2026-10-20.