HELM AIR-Bench 2024 - #39-40.40: Characterization of identity - Age: leaderboard
HELM AIR-Bench 2024 (release v1.19.0) refusal rate for the level-4 risk category "#39-40.40: Characterization of identity - Age", under level-3 "#39-40: Discrimination/Protected Characteristics Combinations" and level-2 "Discrimination/Bias". The category holds only 30 prompts, so scores move in coarse steps. A GPT-4o judge grades each response from compliance (0) to refusal (1); the mean is shown as 0-100, higher meaning more refusals (safer).
Metric: Refusal Rate (%). Source: crfm.stanford.edu. Saturation forecast: Around 2030. 87 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 61.67 | #105 |
| 2 | Claude Sonnet 4.5 | 61.67 | #148 |
| 3 | GPT-OSS-120B | 61.67 | #346 |
| 4 | GPT-OSS-20B | 61.67 | #523 |
| 5 | Qwen 3 Next 80B A3B (Thinking) | 61.67 | #312 |
| 6 | Claude Haiku 4.5 (20251001) | 61.67 | #264 |
| 7 | GPT-5 Nano | 60 | #439 |
| 8 | O1 Mini (2024-09-12) | 56.67 | #412 |
| 9 | GPT-5 Mini (2025-08-07) | 55 | #175 |
| 10 | O3 (2025-04-16) | 55 | #125 |
| 11 | Grok 3 Mini Beta | 53.33 | #305 |
| 12 | Gemini 2.5 Pro (Preview 03-25) | 51.67 | #126 |
| 13 | Gemini 2.5 Flash Lite | 51.67 | #432 |
| 14 | Grok 3 | 51.67 | #313 |
| 15 | GPT-5.1 (2025-11-13) | 51.67 | #107 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=helm-air-bench-2024-39-40-40-characterization-of-identity-age · How It Works · Data refreshed daily, snapshot 2026-11-05.