HELM AIR-Bench 2024 - #39-40.42: Classification of individuals - Ethnicity: leaderboard
Metric: Refusal Rate (%). Source: crfm.stanford.edu. 87 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5 | 90 |
| 2 | Claude Sonnet 4.5 | 90 |
| 3 | Claude Sonnet 4 (20250514) | 90 |
| 4 | Claude Haiku 4.5 (20251001) | 83.33 |
| 5 | Claude Opus 4 (20250514) | 83.33 |
| 6 | GPT-5 Nano | 80 |
| 7 | Claude 3.5 Sonnet (20241022) | 76.67 |
| 8 | GPT-5 Mini (2025-08-07) | 76.67 |
| 9 | O3 (2025-04-16) | 73.33 |
| 10 | Claude 3.5 Sonnet (20240620) | 70 |
| 11 | O1 (2024-12-17) | 70 |
| 12 | O4 Mini (2025-04-16) | 66.67 |
| 13 | GPT-OSS-120B | 60 |
| 14 | Qwen 3 Next 80B A3B (Thinking) | 60 |
| 15 | Gemini 1.5 Flash (002) | 56.67 |
Interactive version: theaggregate.ai/benchmark?slug=helm-air-bench-2024-39-40-42-classification-of-individuals-ethnicity · How It Works · Data refreshed daily, snapshot 2026-09-19.