HELM AIR-Bench 2024 - #39-40.49: Classification of individuals - Geographic region: leaderboard

Metric: Refusal Rate (%). Source: crfm.stanford.edu. 87 models tracked.

Top models

#ModelScore
1GPT-5 Nano46.67
2GPT-543.33
3Claude Sonnet 4.536.67
4Claude Sonnet 4 (20250514)35
5Grok 431.67
6Claude Haiku 4.5 (20251001)26.67
7Qwen 3 Next 80B A3B (Thinking)25
8Claude 3.5 Sonnet (20241022)23.33
9Gemini 1.5 Flash (002)23.33
10GPT-5 Mini (2025-08-07)21.67
11Gemini 1.5 Pro (002)18.33
12O4 Mini (2025-04-16)18.33
13Claude Opus 4 (20250514)18.33
14Gemini 2.5 Pro (Preview 03-25)16.67
15Gemini 3 Pro (Preview)16.67

Interactive version: theaggregate.ai/benchmark?slug=helm-air-bench-2024-39-40-49-classification-of-individuals-geographic-region · How It Works · Data refreshed daily, snapshot 2026-09-19.