Person Identification on 10 public figures cross-identity manipulation 90 adversarial pairings (test)
54.4Targeted ASRGrok 4.2
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Grok 4.22026.05 | 54.4 | 95.6 | |
| Qwen 3.6 Plus2026.05 | 48.9 | 87.8 | |
| Llama 4 Maverick2026.05 | 35.6 | 95.6 | |
| Gemini 3.1 Pro2026.05 | 22.2 | 84.4 |