ATBench - Real-World Harm Diagnosis: leaderboard
Metric: Fine-grained diagnosis accuracy (%) of the real-world-harm label (ATBench taxonomy) over the 497 unsafe ATBench trajectories, with the AgentDoG diagnosis template; higher is better. Source: arxiv.org. Saturation forecast: Around 2037. 13 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.4 | 30.2 |
| 2 | GPT-5.2 | 26.8 |
| 3 | Qwen 3 235B A22B 2507 Instruct | 26.6 |
| 4 | QwQ-32B | 22.9 |
| 5 | Qwen 3 4B 2507 Instruct | 21.2 |
| 6 | Gemini 3.1 Pro (Preview) | 18.5 |
| 7 | Qwen 3 4B | 18.3 |
| 8 | Llama 3.1 8B Instruct | 15.5 |
| 9 | Qwen 2.5 7B Instruct | 15.5 |
| 10 | Gemini 3 Flash | 15 |
| 11 | Qwen 3.5 4B | 8.2 |
| 12 | Qwen 3.5 397B A17B | 6.8 |
Interactive version: theaggregate.ai/benchmark?slug=atbench-real-world-harm-diagnosis · How It Works · Data refreshed daily, snapshot 2026-10-07.