HELM Safety - Bbq: leaderboard
The "Bbq" subtask of HELM Safety. HELM Safety: Measures model robustness, truthfulness, calibration, bias, harmfulness, jailbreak resistance, or alignment-relevant behavior.
Metric: BBQ accuracy. Source: crfm.stanford.edu. Source: storage.googleapis.com. Saturation forecast: Estimated already saturated. 87 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4 (20250514) | 99.3 | #154 |
| 2 | Claude Sonnet 4.5 | 98.9 | #142 |
| 3 | GPT-OSS-120B | 98.5 | #341 |
| 4 | Gemini 3 Pro (Preview) | 98.4 | #66 |
| 5 | O3 (2025-04-16) | 97.9 | #121 |
| 6 | Claude Sonnet 4 (20250514) | 97.9 | #213 |
| 7 | Gemini 2.5 Flash (Preview 04-17) | 97.7 | #237 |
| 8 | GPT-5 Nano | 97.6 | #425 |
| 9 | O1 (2024-12-17) | 97.3 | #149 |
| 10 | Claude Opus 4 (20250514) (Thinking) | 97.1 | #154 (Claude Opus 4 (20250514)) |
| 11 | O1 Mini (2024-09-12) | 96.9 | #403 |
| 12 | GPT-5 | 96.8 | #98 |
| 13 | GPT-OSS-20B | 96.7 | #518 |
| 14 | DeepSeek V3 | 96.7 | #321 |
| 15 | Grok 3 Mini Beta | 96.7 | #296 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=helm-safety-bbq · How It Works · Data refreshed daily, snapshot 2026-10-20.