BlueBench - Bias: leaderboard
BlueBench bias task using BBQ-style ambiguous and disambiguated QA to measure whether models answer without relying on protected-class stereotypes.
Metric: Score (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 18 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4.1 | 97.98 | #248 |
| 2 | O3 Mini | 97.98 | #272 |
| 3 | Mistral Large | 96.97 | #710 |
| 4 | Llama 3.3 70B Instruct | 95.96 | #534 |
| 5 | GPT-4.1 Mini | 94.95 | #355 |
| 6 | O4 Mini | 94.95 | #180 |
| 7 | O1 | 94.95 | #183 |
| 8 | GPT-4o | 93.94 | #343 |
| 9 | Mistral Medium 3 | 93.94 | #500 |
| 10 | GPT-4.1 Nano | 81.82 | #731 |
| 11 | Granite 3.3 8B Instruct | 76.77 | #825 |
| 12 | Pixtral-12B | 72.73 | #808 |
| 13 | Llama 3.2 3B Instruct | 60.61 | #1352 |
| 14 | Llama 3.2 1B Instruct | 48.48 | #1539 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=bluebench-bias · How It Works · Data refreshed daily, snapshot 2026-10-20.