GenderBench - GEST Stereotype Rate: leaderboard
The model is shown first-person quotes from the GEST dataset that express gender stereotypes and asked whether a man or a woman said them; the number is the excess of stereotypical attributions, normalized as max(0, x). The creative GEST board asks for stories instead.
Metric: Excess of stereotypical gender attributions of GEST quotes, max(0, x) (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 | 0.11 | #4246 |
| 2 | Mistral 7B Instruct (v0.3) | 0.14 | #6653 |
| 3 | Mistral Small 3 | 0.17 | #1777 |
| 4 | GPT-4o | 0.24 | #741 |
| 5 | GPT-4o Mini | 0.42 | #1623 |
| 6 | Phi-4 | 0.42 | #1913 |
| 7 | Gemma 2 9B (IT) | 0.48 | #2262 |
| 8 | Gemini 2.0 Flash Lite | 0.54 | #1032 |
| 9 | Gemma 2 27B (IT) | 0.56 | #1782 |
| 10 | Claude 3.5 Haiku | 0.58 | #1553 |
| 11 | Llama 3.3 70B Instruct | 0.64 | #1357 |
| 12 | Gemini 2.0 Flash | 0.69 | #704 |
Interactive version: theaggregate.ai/benchmark?slug=genderbench-gest-stereotype-rate · How It Works · Data refreshed daily, snapshot 2026-11-13.