Risk Identification on IS-Bench
69.9Step AccuracyGPT-5.1
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| GPT-5.12026.05 | 69.9 | 29.1 | 64.2 | 38.9 | |
| Qwen-3-VL-32BModel size=32B2026.05 | 66.7 | 27.8 | 70.8 | 41 | |
| EMBGUARD-4BModel size=4B2026.05 | 63.1 | 25.7 | 71.7 | 38.3 | |
| Gemini-2.5-Pro2026.05 | 49.9 | 22.2 | 88.2 | 40.7 | |
| EMBGUARD-2BModel size=2B2026.05 | 45.7 | 19.1 | 76.7 | 30.5 |