PHANTOM - FC ATTACK ASR: leaderboard
Metric: Reported aggregate attack success rate (%) for FC ATTACK attacks on PHANTOM; lower is better for victim-model safety robustness. Paper §3.4 and Appendix C describe one fixed subset per attack reused across all models (nominal 1,100 attacks, 20 per subcategory), judged with Abel-24-HarmClassifier; empty API responses count as failed jailbreaks. Preserve reported aggregate rather than reconstructing a count from the nominal denominator.. Source: arxiv.org. 15 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemma 4 26B A4B (IT) | 18.91 |
| 2 | Gemini 3.1 Pro (Preview) | 37 |
| 3 | GPT-5.4 | 61 |
| 4 | Ministral-3-14B-Instruct-2512 | 61.27 |
| 5 | Claude Opus 4.7 | 63.45 |
| 6 | Claude Opus 4.8 | 67.45 |
| 7 | Qwen 3.5 27B | 68.27 |
| 8 | Claude Opus 4.6 | 70.82 |
| 9 | GPT-5.5 | 71.36 |
| 10 | Qwen 3 VL 30B A3B Instruct | 73.45 |
| 11 | Qwen 3.6 27B | 77.27 |
Interactive version: theaggregate.ai/benchmark?slug=phantom-fc-attack-asr · How It Works · Data refreshed daily, snapshot 2026-10-09.