ATBench - Risk Source Diagnosis: leaderboard
Metric: Fine-grained diagnosis accuracy (%) of the risk-source label (ATBench taxonomy) over the 497 unsafe ATBench trajectories, with the AgentDoG diagnosis template; higher is better. Source: arxiv.org. Saturation forecast: Around January 2028. 13 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.4 | 33.6 |
| 2 | GPT-5.2 | 29.5 |
| 3 | Gemini 3.1 Pro (Preview) | 24.8 |
| 4 | Gemini 3 Flash | 18.4 |
| 5 | QwQ-32B | 15.8 |
| 6 | Qwen 3.5 397B A17B | 7.7 |
| 7 | Qwen 3 235B A22B 2507 Instruct | 7 |
| 8 | Qwen 3.5 4B | 6.6 |
| 9 | Llama 3.1 8B Instruct | 6.2 |
| 10 | Qwen 2.5 7B Instruct | 5.3 |
| 11 | Qwen 3 4B | 4.4 |
| 12 | Qwen 3 4B 2507 Instruct | 1 |
Interactive version: theaggregate.ai/benchmark?slug=atbench-risk-source-diagnosis · How It Works · Data refreshed daily, snapshot 2026-10-07.