ATBench - Real-World Harm Diagnosis: leaderboard
ATBench (2026) tests LLMs and guard models as safety judges of long-horizon agent tool-use trajectories built from a taxonomy of risk sources, failure modes and real-world harms. This board is the accuracy in percent of naming the real-world harm of each unsafe trajectory; higher is better.
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 2035. 13 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | AgentDoG-Qwen3-4B | 40.6 | |
| 2 | GPT-5.4 | 30.2 | #135 |
| 3 | GPT-5.2 | 26.8 | #185 |
| 4 | Qwen 3 235B A22B 2507 Instruct | 26.6 | #616 |
| 5 | QwQ-32B | 22.9 | #907 |
| 6 | Qwen 3 4B 2507 Instruct | 21.2 | #2221 |
| 7 | Gemini 3.1 Pro (Preview) | 18.5 | #99 |
| 8 | Qwen 3 4B | 18.3 | #2348 |
| 9 | Llama 3.1 8B Instruct | 15.5 | #4600 |
| 10 | Qwen 2.5 7B Instruct | 15.5 | #2723 |
| 11 | Gemini 3 Flash | 15 | #175 |
| 12 | Qwen 3.5 4B | 8.2 | #1241 |
| 13 | Qwen 3.5 397B A17B | 6.8 | #264 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=atbench-real-world-harm-diagnosis · How It Works · Data refreshed daily, snapshot 2026-11-21.