LiveSecBench: leaderboard
Dynamic live safety benchmark for large language models across ethics, legality, privacy, factuality, and psychological health.
Metric: Overall Score (%). Source: livesecbench.intokentech.cn. Saturation forecast: Estimated already saturated. 43 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Haiku 4.5 | 91.43 | #280 |
| 2 | Claude Sonnet 4.6 | 85.97 | #90 |
| 3 | GPT-5.2 | 84.72 | #110 |
| 4 | Qwen 3.5 Plus (2026-02-15) | 84.34 | |
| 5 | Qwen 3.5 397B A17B | 81.52 | #146 |
| 6 | Spark X2 | 79.18 | #364 |
| 7 | Kimi K2.5 | 74.79 | #144 |
| 8 | Seed-1.6 | 70.83 | #265 |
| 9 | Qwen 3 235B A22B | 69.23 | #313 |
| 10 | MiniMax-M2 | 66.69 | #317 |
| 11 | GPT-OSS-120B | 66.63 | #341 |
| 12 | Seed 2.0 Pro | 63.04 | #118 |
| 13 | MiniMax-M2.5 | 61.65 | #303 |
| 14 | Gemini 3.1 Pro (Preview) | 58.16 | #58 |
| 15 | MiMo-V2-Flash | 57.23 | #372 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=livesecbench · How It Works · Data refreshed daily, snapshot 2026-10-20.