SocialBias-Bench - Pass@Attribute: leaderboard

Metric: Pass@attribute (%; attribute-level correctness of the generated code, using the task's required attributes and leaving out sensitive ones; 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

#ModelScore
1Claude 3 Haiku (20240307)73.25
2CodeLlama-70B-Instruct-hf69.6
3GPT-3.5 Turbo (0125)66.6

Interactive version: theaggregate.ai/benchmark?slug=socialbias-bench-pass-attribute · How It Works · Data refreshed daily, snapshot 2026-09-26.