GenderBench - Direct Stereotype Agreement: leaderboard
The model is asked directly whether it agrees with stereotypical statements about men and women; the number is the share it agrees with. It tests stated beliefs, unlike the probes that infer bias from decisions.
Metric: Rate of agreeing with stereotypical statements about genders (0-1 normalized, lower is better). Source: genderbench.readthedocs.io. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Llama 3.1 8B Instruct | 0.02 | #4246 |
| 2 | Claude 3.5 Haiku | 0.03 | #1553 |
| 3 | Gemma 2 9B (IT) | 0.03 | #2262 |
| 4 | Phi-4 | 0.03 | #1913 |
| 5 | Mistral Small 3 | 0.03 | #1777 |
| 6 | Gemma 2 27B (IT) | 0.04 | #1782 |
| 7 | Gemini 2.0 Flash Lite | 0.04 | #1032 |
| 8 | Llama 3.3 70B Instruct | 0.04 | #1357 |
| 9 | Gemini 2.0 Flash | 0.05 | #704 |
| 10 | GPT-4o | 0.05 | #741 |
| 11 | Mistral 7B Instruct (v0.3) | 0.05 | #6653 |
| 12 | GPT-4o Mini | 0.09 | #1623 |
Interactive version: theaggregate.ai/benchmark?slug=genderbench-direct-stereotype-agreement · How It Works · Data refreshed daily, snapshot 2026-11-13.