Safety-Flag: leaderboard
Metric: Macro-F1 (%; mean over the seven source benchmarks of the macro-F1 of flag / do-not-flag verdicts on 198-200 class-balanced items from each of BeaverTails, XSTest, Ethics, WildGuard, Aegis, ToxiChat and ToxiGen, recast as one flag / do-not-flag decision with harmonised source labels; one standardized two-option prompt with a JSON verdict and confidence; greedy decoding). Source: arxiv.org. Saturation forecast: Around December 2026. 9 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.4 Mini | 85.3 |
| 2 | GPT-4.1 Mini | 84.3 |
| 3 | Gemma 2 9B (IT) | 82.4 |
| 4 | DeepSeek R1 Distill Llama 8B | 81.3 |
| 5 | Qwen 2.5 7B Instruct | 79.3 |
| 6 | Mistral 7B Instruct (v0.3) | 73.3 |
| 7 | OLMo-2-1124-7B-Instruct | 63.9 |
| 8 | Llama 3.1 8B Instruct | 44.3 |
Interactive version: theaggregate.ai/benchmark?slug=safety-flag · How It Works · Data refreshed daily, snapshot 2026-09-26.