TRACE (LRM Safety) - Prompt Risk Category: leaderboard
TRACE (Wu et al., 2026) tests guardrail models on 5,000 prompt, reasoning-trace and final-response triples from four large reasoning models in English and Chinese, with safety labels and evidence spans from a majority of three LLM annotators. This board scores the classification of unsafe prompts into nine risk categories by guardrails that accept custom categories. Macro F1 in percent; higher is better.
Metric: Macro F1 (%; classification of unsafe prompts into nine risk categories by guardrails that support custom categories, TRACE English and Chinese). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | gpt-oss-safeguard-20b | 62.25 | |
| 2 | YuFeng-XGuard-Reason-8B | 53.24 | |
| 3 | YuFeng-XGuard-Reason-0.6B | 52.62 | |
| 4 | Llama-3.1-Nemotron-Safety-Guard-8B-v3 | 27.13 | |
| 5 | PolyGuard-Qwen | 16.63 | |
| 6 | PolyGuard-Qwen-Smol | 7.68 |
Interactive version: theaggregate.ai/benchmark?slug=trace-lrm-safety-prompt-risk-category · How It Works · Data refreshed daily, snapshot 2026-11-12.