GuardZoo (Zero-Shot) - Safe/Unsafe Accuracy: leaderboard

Metric: Binary safe/unsafe accuracy (%; GuardZoo in-domain test split, 20% of the Aegis 2.0, WildGuardMix, BeaverTails, S-Eval, PolicyGuard-Bench, ASSE-Bench and R-Judge samples, relabeled; zero-shot guard models). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 9 models tracked.

Top models

#ModelScore
1Qwen3Guard-4B77
2Qwen3Guard-8B76.14
3Llama Guard 3 8B70
4ShieldGemma 2B68.09
5LlamaGuard 2 8B67.71
6Qwen3Guard-0.6B66.5
7ShieldGemma 9B63.74
8ShieldGemma 27B52.07
9Llama Guard 3 1B45.13

Interactive version: theaggregate.ai/benchmark?slug=guardzoo-zero-shot-safe-unsafe-accuracy · How It Works · Data refreshed daily, snapshot 2026-09-26.