SEAL - Fortress: leaderboard
Safety evaluation in high-stakes domains including CBRNE threats, political violence, and financial crime, measuring both refusal of harmful requests and avoidance of over-refusing legitimate queries.
Metric: Score. Source: scale.com. 71 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DeepSeek R1 | 74.39 | #251 |
| 2 | GLM 4.5 Air | 63.18 | #392 |
| 3 | Gemini 2.5 Pro (Preview 06-05) | 61.68 | #106 |
| 4 | Qwen 3 235B A22B | 61.39 | #313 |
| 5 | DeepSeek V3.1 | 60.55 | #264 |
| 6 | GLM-4.5 | 59.58 | #275 |
| 7 | GPT-4.1 Mini | 59.19 | #355 |
| 8 | Qwen 2.5 72B | 56.43 | #499 |
| 9 | Mixtral 8x22B | 56.06 | #766 |
| 10 | Kimi K2 | 55.47 | #243 |
| 11 | Gemini 2.5 Pro (Preview 03-25) | 54.89 | #123 |
| 12 | Gemini 1.5 Pro | 53.86 | #445 |
| 13 | GPT-4.1 | 53.02 | #248 |
| 14 | Gemini 3.1 Flash Lite (Preview) | 51.05 | #194 |
| 15 | Gemini 1.5 Flash | 50.61 | #618 |
Interactive version: theaggregate.ai/benchmark?slug=seal-fortress · How It Works · Data refreshed daily, snapshot 2026-10-20.