MIR-SafetyBench - Spatial Juxtaposition: leaderboard
Metric: Attack Success Rate (%; lower is safer). Source: arxiv.org. Saturation forecast: Around December 2026. 19 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.1 | 17.47 |
| 2 | Gemini 3 Pro (Preview) | 41.78 |
| 3 | Gemini 2.5 Pro | 52.05 |
| 4 | GPT-4o Mini | 56.16 |
| 5 | QVQ-72B-Preview | 72.26 |
| 6 | Gemini 2.5 Flash | 74.32 |
| 7 | InternVL3-8B | 75.68 |
| 8 | InternVL3-38B | 76.71 |
| 9 | GPT-4o | 77.05 |
| 10 | InternVL3-78B | 78.08 |
| 11 | GLM-4.1V-9B (Thinking) | 88.01 |
| 12 | Qwen 2.5 VL 32B Instruct | 89.73 |
Interactive version: theaggregate.ai/benchmark?slug=mir-safetybench-spatial-juxtaposition · How It Works · Data refreshed daily, snapshot 2026-09-25.