GenderBench - DiscrimEval Decision Gap: leaderboard
The model decides yes/no in high-stakes scenarios from Anthropic's DiscrimEval (loans, hiring, housing) with only the person's gender varied; the number is the largest gap in favourable-decision rate between genders. This is GenderBench's gender-only re-run, not the original DiscrimEval logit score.
Metric: Largest gender gap in favourable decisions on DiscrimEval scenarios (0-1 normalized, lower is better). Source: genderbench.readthedocs.io. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o | 0.01 | #741 |
| 2 | Gemini 2.0 Flash Lite | 0.01 | #1032 |
| 3 | Mistral 7B Instruct (v0.3) | 0.01 | #6653 |
| 4 | Llama 3.3 70B Instruct | 0.01 | #1357 |
| 5 | GPT-4o Mini | 0.02 | #1623 |
| 6 | Gemini 2.0 Flash | 0.02 | #704 |
| 7 | Phi-4 | 0.02 | #1913 |
| 8 | Mistral Small 3 | 0.04 | #1777 |
| 9 | Gemma 2 27B (IT) | 0.04 | #1782 |
| 10 | Gemma 2 9B (IT) | 0.04 | #2262 |
| 11 | Claude 3.5 Haiku | 0.06 | #1553 |
| 12 | Llama 3.1 8B Instruct | 0.08 | #4246 |
Interactive version: theaggregate.ai/benchmark?slug=genderbench-discrimeval-decision-gap · How It Works · Data refreshed daily, snapshot 2026-11-13.