MME-Safety - Relevance Rate: leaderboard
Metric: Relevance rate (%; share of final answers to 3,801 manually verified image-text pairs with harmful intent across nine risk categories and nine image/text stealth combinations, zero-shot; GPT-4o judge whose response is logically and semantically relevant to the query, refusals included). Source: arxiv.org. Saturation forecast: Estimated already saturated. 17 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | O4 Mini | 90.94 |
| 2 | Claude Sonnet 4 (Thinking) | 88.53 |
| 3 | Claude 3.7 Sonnet | 84.58 |
| 4 | Qwen 3 VL 235B A22B Instruct | 84.24 |
| 5 | Gemini 3 Pro | 83.17 |
| 6 | Qwen 3 VL 32B (Thinking) | 82.58 |
| 7 | Qwen 3 VL 235B A22B (Thinking) | 82.22 |
| 8 | Claude 3.7 Sonnet (Thinking) | 82.14 |
| 9 | Qwen 3 VL 8B (Thinking) | 81.17 |
Interactive version: theaggregate.ai/benchmark?slug=mme-safety-relevance-rate · How It Works · Data refreshed daily, snapshot 2026-09-26.