HELM AIR-Bench 2024 - #14.14: Color: leaderboard
HELM AIR-Bench 2024 (release v1.19.0) refusal rate for the level-4 risk category "#14.14: Color", under level-3 "#14: Hate Speech (Inciting/Promoting/Expressing hatred)" and level-2 "Hate/Toxicity" of the regulation-derived AIR 2024 taxonomy. The category holds only 15 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: Estimated already saturated. 87 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 100 | #105 |
| 2 | Llama 3.1 8B Instruct | 100 | #1017 |
| 3 | Claude Sonnet 4.5 | 100 | #148 |
| 4 | GPT-4.1 Mini | 100 | #363 |
| 5 | GPT-OSS-120B | 100 | #346 |
| 6 | Claude 3.5 Sonnet (20241022) | 100 | #302 |
| 7 | GPT-5 Nano | 100 | #439 |
| 8 | Gemini 2.0 Flash | 100 | #351 |
| 9 | Gemini 3 Pro (Preview) | 100 | #71 |
| 10 | Gemini 1.5 Pro (002) | 100 | #368 |
| 11 | DeepSeek R1 | 100 | #262 |
| 12 | Gemini 1.5 Flash (002) | 100 | #587 |
| 13 | Claude 3.5 Sonnet (20240620) | 100 | #308 |
| 14 | GPT-OSS-20B | 100 | #523 |
| 15 | Gemini 2.5 Pro (Preview 03-25) | 100 | #126 |
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-2024-14-14-color · How It Works · Data refreshed daily, snapshot 2026-11-05.