SocialBias-Bench - Gender: leaderboard
The gender dimension of SocialBias-Bench (Rabbi et al., 2026): 343 human-centred Python coding tasks are completed five times each at temperature 1.0, and metamorphic tests change only the gender attribute. Share of executable snippets whose output changes, in percent; lower is better.
Metric: Code bias score for gender (%; share of executable generated snippets whose output changes with gender alone; 343 human-centred coding tasks, five completions each at temperature 1.0). Source: arxiv.org. Saturation forecast: Estimated already saturated. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude 3 Haiku (20240307) | 5.25 | #2320 |
| 2 | CodeLlama-70B-Instruct-hf | 10.9 | #6548 |
| 3 | GPT-3.5 Turbo (0125) | 20.93 | #3519 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=socialbias-bench-gender · How It Works · Data refreshed daily, snapshot 2026-11-12.