HELM AIR-Bench 2024 - #6.25: Financial: leaderboard
HELM AIR-Bench 2024 (release v1.19.0) refusal rate for the level-4 risk category "#6.25: Financial", under level-3 "#6: Advice in Heavily Regulated Industries" and level-2 "Operational Misuses" 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. 87 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude 3.5 Sonnet (20241022) | 43.33 | #302 |
| 2 | Claude Haiku 4.5 (20251001) | 33.33 | #264 |
| 3 | GPT-OSS-120B | 26.67 | #346 |
| 4 | Claude 3 Haiku (20240307) | 23.33 | #780 |
| 5 | Claude 3 Sonnet (20240229) | 23.33 | #694 |
| 6 | Llama 3.1 8B Instruct | 20 | #1017 |
| 7 | GPT-OSS-20B | 16.67 | #523 |
| 8 | Claude 3.5 Sonnet (20240620) | 13.33 | #308 |
| 9 | Llama 3 8B Instruct | 13.33 | #1116 |
| 10 | Llama 3.1 405B Instruct | 13.33 | #470 |
| 11 | Claude 3.7 Sonnet (20250219) | 13.33 | #209 |
| 12 | GPT-3.5 Turbo (0125) | 13.33 | #975 |
| 13 | Claude Sonnet 4 (20250514) | 13.33 | #223 |
| 14 | GPT-3.5 Turbo (0613) | 13.33 | #1106 |
| 15 | GPT-3.5 Turbo (1106) | 13.33 | #952 |
Interactive version: theaggregate.ai/benchmark?slug=helm-air-bench-2024-6-25-financial · How It Works · Data refreshed daily, snapshot 2026-11-05.