MME-Safety: leaderboard
MME-Safety (Shi et al., 2026) evaluates the safety of multimodal language models on 3,801 manually verified image-text pairs with harmful intent, balanced over nine risk categories and nine combinations of image and text stealth level, scoring zero-shot final answers with a GPT-4o judge. The overall score in percent averages relevance, one minus harmful rate times normalised harm severity, and refusal rate; higher is better.
Metric: Overall safety score (%; mean of relevance rate, 1 - harmful rate x normalised average harm severity, and refusal rate, all on 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). Source: arxiv.org. Saturation forecast: Estimated already saturated. 17 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4 (Thinking) | 86.24 | #397 (Claude Sonnet 4) |
| 2 | Qwen 3 VL 235B A22B Instruct | 85.11 | #521 |
| 3 | O4 Mini | 84.78 | #354 |
| 4 | Claude 3.7 Sonnet | 84.42 | #455 |
| 5 | Claude 3.7 Sonnet (Thinking) | 82.17 | #455 (Claude 3.7 Sonnet) |
| 6 | Qwen 3 VL 235B A22B (Thinking) | 81.16 | #451 |
| 7 | Qwen 3 VL 32B (Thinking) | 80.49 | #557 |
| 8 | Qwen 3 VL 8B (Thinking) | 77.33 | #887 |
| 9 | Gemini 3 Pro | 75.55 | #135 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=mme-safety · How It Works · Data refreshed daily, snapshot 2026-11-12.