E-commerce adversarial robustness on EVADE-Bench
58.9Robustness ScoreYuvion VL
Evaluation results
| Method | Links | |
|---|---|---|
| Yuvion VLModel Scale=32B, Decoding Mode=Reasoning2026.06 | 58.9 | |
| Yuvion VLModel Scale=8B, Decoding Mode=Reasoning2026.06 | 54 | |
| K2.5Model Scale=≈1T2026.06 | 53.4 | |
| Yuvion VLModel Scale=32B, Decoding Mode=Instruct2026.06 | 53.1 | |
| GPT-5.4Model Scale=≈1.5T2026.06 | 50.8 | |
| Yuvion VLModel Scale=8B, Decoding Mode=Instruct2026.06 | 50.4 | |
| Qwen3.5-PlusModel Scale=397B2026.06 | 50 | |
| Qwen3-VLModel Scale=32B, Decoding Mode=Instruct2026.06 | 47.4 | |
| Qwen3-VLModel Scale=8B, Decoding Mode=Instruct2026.06 | 47.2 | |
| Opus-4.6Model Scale=≈800B2026.06 | 46.9 | |
| Qwen3-VLModel Scale=8B, Decoding Mode=Thinking2026.06 | 46.6 | |
| Qwen3-VLModel Scale=32B, Decoding Mode=Thinking2026.06 | 45.2 |