HarmVideoBench - Clip-Internal Meaning: leaderboard
HarmVideoBench (2026) asks multimodal models four-option questions about 1,379 real harmful videos at three levels of understanding; API models watch the video with audio and open models see 12 key frames with the transcript. This board is accuracy in percent on questions about the meaning implied within the clip; chance is 25 and higher is better.
Metric: Clip-Internal Meaning accuracy (%) on four-option multiple-choice questions about 1,379 real harmful videos (one question per video per level), temperature 0 greedy decoding; API models take the video and audio natively, open models 12 sampled key frames plus the transcript; chance 25; questions about the implied meaning within the clip; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 19 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 84.6 | #94 |
| 2 | GPT-5.5 | 83.4 | #49 |
| 3 | Claude Opus 4.7 | 82.9 | #79 |
| 4 | Qwen 3 VL 235B A22B (Thinking) | 82.6 | #451 |
| 5 | Claude Sonnet 4.6 | 82.5 | #161 |
| 6 | Qwen 3 VL 235B A22B Instruct | 82.3 | #522 |
| 7 | Qwen 3 VL 32B (Thinking) | 82.1 | #559 |
| 8 | Qwen 3 VL 32B Instruct | 81.6 | #585 |
| 9 | Qwen 3 VL 30B A3B (Thinking) | 81.4 | #712 |
| 10 | Qwen 3 VL 8B (Thinking) | 79.4 | #887 |
| 11 | Qwen 3 VL 30B A3B Instruct | 78.9 | #871 |
| 12 | MiniCPM-V-4.6 | 72.4 | #3525 |
| 13 | LFM2.5-VL-1.6B | 69.3 | #3872 |
| 14 | GLM-4.5V (Non-reasoning) | 67.8 | #744 (GLM-4.5V) |
| 15 | Qwen 3 VL 8B Instruct | 63.7 | #981 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=harmvideobench-clip-internal-meaning · How It Works · Data refreshed daily, snapshot 2026-11-13.