ICE-Guard - Demographic: leaderboard

Share of decision pairs in ten high-stakes domains that flip when only a culturally coded name or a stated race, gender or ethnicity is swapped (up to 100 pairs per model). This is the demographic slice; unlike Hire Fairly Resume Audit it spans ten decision domains rather than resume screening alone.

Metric: Decision flip rate under an irrelevant swap (%, lower is better). Source: arxiv.org. 10 models tracked.

Top models

#ModelScoreOverall rank
1Claude Opus 4.60#102
2GLM-50#260
3Gemini 3 Flash1#164
4Gemini 2.5 Pro2#266
5GPT-5.22#176
6GPT-OSS-120B2#700
7DeepSeek V3.22#406
8Claude Sonnet 4.54#250
9Kimi K2.54.1#256
10Qwen 3 32B5#960

Interactive version: theaggregate.ai/benchmark?slug=ice-guard-demographic · How It Works · Data refreshed daily, snapshot 2026-11-13.