Phare - Harm Resistance: leaderboard
Giskard’s Phare multilingual safety probe (English, French, Spanish), Harm Resistance module: whether it declines requests for harmful content. The metric is a resistance score, so a higher number means the model held the line more often.
Metric: Score (%). Source: phare.giskard.ai. Saturation forecast: Estimated already saturated. 71 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.6 | 100 | #63 |
| 2 | Claude Sonnet 4.6 | 99.93 | #90 |
| 3 | Claude Haiku 4.5 | 99.93 | #280 |
| 4 | Claude Sonnet 5 | 99.59 | #57 |
| 5 | Claude Sonnet 4.5 | 99.05 | #142 |
| 6 | GPT-5.5 | 98.92 | #31 |
| 7 | GPT-5 Mini | 98.29 | #186 |
| 8 | Claude Opus 4.5 | 98.25 | #83 |
| 9 | GPT-5 Nano | 97.41 | #425 |
| 10 | Kimi K2.5 | 97.2 | #144 |
| 11 | Kimi K2.6 | 97 | #104 |
| 12 | GPT-5 | 96.97 | #98 |
| 13 | Gemini 3.1 Flash Lite | 96.94 | #191 |
| 14 | GPT-5.1 | 96.92 | #135 |
| 15 | GPT-5.2 | 96.88 | #110 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=phare-harm-resistance · How It Works · Data refreshed daily, snapshot 2026-10-20.