HarmVideoBench - Observable Evidence: leaderboard
Metric: Observable Evidence 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 what is directly visible or audible in the clip; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 19 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 96 |
| 2 | GPT-5.5 | 95.4 |
| 3 | Qwen 3 VL 235B A22B (Thinking) | 95.1 |
| 4 | Qwen 3 VL 235B A22B Instruct | 94.8 |
| 5 | Claude Opus 4.7 | 94.7 |
| 6 | Qwen 3 VL 32B (Thinking) | 94.7 |
| 7 | Claude Sonnet 4.6 | 94.4 |
| 8 | Qwen 3 VL 32B Instruct | 94.3 |
| 9 | Qwen 3 VL 30B A3B (Thinking) | 94 |
| 10 | Qwen 3 VL 30B A3B Instruct | 91.5 |
| 11 | Qwen 3 VL 8B (Thinking) | 91.2 |
| 12 | LFM2.5-VL-1.6B | 79.8 |
| 13 | GLM-4.5V (Non-reasoning) | 77.7 |
| 14 | Qwen 3 VL 8B Instruct | 73.3 |
Interactive version: theaggregate.ai/benchmark?slug=harmvideobench-observable-evidence · How It Works · Data refreshed daily, snapshot 2026-09-29.