Bias Evaluation on BBQ (Steerability Metrics)
1.84Steerability ScoreDeepSeek V3.2
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| DeepSeek V3.2Reasoning=low2026.02 | 1.84 | 0.18 | 91.1 | 21.4 | 0 | |
| Qwen3-235BReasoning=low2026.02 | 1.24 | 0.32 | 83.9 | 35.7 | 2.1 | |
| DeepSeek V3.2Reasoning=off2026.02 | 1 | 0.37 | 80.4 | 39.3 | 2.2 | |
| Llama 3.3 70BReasoning=none2026.02 | 0.89 | 0.36 | 67.9 | 35.7 | 0 | |
| Qwen3-235BReasoning=off2026.02 | 0.86 | 0.34 | 76.8 | 32.1 | 4.7 | |
| Llama 3.3 70BReasoning=before2026.02 | 0.71 | 0.31 | 66.1 | 28.6 | 0 | |
| GPT-5.2Reasoning=low2026.02 | 0.64 | 0.61 | 48.2 | 46.4 | 40.7 | |
| GPT-5.2Reasoning=off2026.02 | 0.48 | 0.41 | 58.9 | 21.4 | 72.7 | |
| Grok 4.1 FastReasoning=off2026.02 | 0.35 | 0.4 | 41.1 | 17.9 | 8.7 |