HarmVideoBench: leaderboard
Large vision-language models (LVLMs) have recently shown immense potential in automated content moderation, sparking growing interest in developing harmful-video benchmarks. However, we identify two primary limitations in existing works: 1) The multi-layered characteristics of harmful videos are overlooked. Existing benchmarks predominantly formulate evaluation as a binary classification task, fai
Metric: Macro Avg. (self-reported). Source: benchmarklist.com. Saturation forecast: Around June 2027. 21 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 82.1 | #58 |
| 2 | GPT-5.5 | 80.9 | #31 |
| 3 | Claude Opus 4.7 | 80.3 | #49 |
| 4 | Qwen 3 VL 235B A22B (Thinking) | 80.1 | #238 (Qwen 3 VL 235B A22B) |
| 5 | Claude Sonnet 4.6 | 79.9 | #90 |
| 6 | Qwen 3 VL 235B A22B Instruct | 79.7 | #273 |
| 7 | Qwen 3 VL 32B (Thinking) | 79.6 | #294 (Qwen 3 VL 32B) |
| 8 | Qwen 3 VL 32B Instruct | 79 | #283 |
| 9 | Qwen 3 VL 30B A3B (Thinking) | 78.9 | #351 (Qwen 3 VL 30B A3B) |
| 10 | Qwen 3 VL 8B (Thinking) | 76.7 | |
| 11 | Qwen 3 VL 30B A3B Instruct | 76.3 | #384 |
| 12 | LFM2.5-VL-1.6B | 66.8 | #854 |
| 13 | GLM-4.5V | 65.5 | #346 |
| 14 | Qwen 3 VL 8B Instruct | 61.7 | #420 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=harmvideobench · How It Works · Data refreshed daily, snapshot 2026-10-20.