Risk Source Prediction on ATBench
52AccuracyAgentDoG
Evaluation Results
| Method | Links | |
|---|---|---|
| AgentDoGBase Model=Qwen3-4B2026.05 | 52 | |
| AgentDoGBase Model=Llama3.1-8B-Instruct2026.05 | 51.6 | |
| TRACESBase Model=Llama3.1-8B-Instruct2026.05 | 50 | |
| TRACESBase Model=Qwen3-4B2026.05 | 48.8 | |
| GPT-5.2Base Model=Original2026.05 | 41.6 | |
| Gemini-3-FlashBase Model=Original2026.05 | 38 | |
| Gemini-3-ProBase Model=Original2026.05 | 36.8 | |
| QWQ-32BBase Model=Original2026.05 | 23.2 | |
| Llama3.1-8BBase Model=Original2026.05 | 6.2 | |
| Qwen3-4BBase Model=Original2026.05 | 4.2 |