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. Status: saturation imminent. 43 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Haiku 4.5 | 91.43 |
| 2 | Claude Sonnet 4.6 | 85.97 |
| 3 | GPT-5.2 | 84.72 |
| 4 | Qwen 3.5 397B A17B | 81.52 |
| 5 | Spark X2 | 79.18 |
| 6 | Kimi K2.5 | 74.79 |
| 7 | Doubao-Seed-1.6 | 70.83 |
| 8 | Qwen 3 235B A22B | 69.23 |
| 9 | MiniMax-M2 | 66.69 |
| 10 | GPT-OSS-120B | 66.63 |
| 11 | Seed 2.0 Pro | 63.04 |
| 12 | MiniMax-M2.5 | 61.65 |
| 13 | Gemini 3.1 Pro (Preview) | 58.16 |
| 14 | MiMo-V2-Flash | 57.23 |
| 15 | LongCat-Flash-Chat | 57.1 |
Interactive version: theaggregate.ai/benchmark?slug=livesecbench · How It Works · Data refreshed daily, snapshot 2026-09-05.