Steering on Principle-based evaluation dataset
8.68G ScoreClaude 4.5-Haiku
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Claude 4.5-HaikuAccess Type=API-based2025.12 | 8.68 | 8.76 | 8.13 | 8.52 | 8.67 | 8.53 | |
| Claude 4.5-SonnetAccess Type=API-based2025.12 | 8.6 | 8.76 | 8.09 | 8.52 | 8.61 | 8.6 | |
| GPT-5Access Type=API-based2025.12 | 8.53 | 8.51 | 7.78 | 8.33 | 8.51 | 8.3 | |
| GPT-5.1Access Type=API-based2025.12 | 8.49 | 8.59 | 7.85 | 8.51 | 8.61 | 8.38 | |
| Claude 3.5-HaikuAccess Type=API-based2025.12 | 8.47 | 8.45 | 7.88 | 8.36 | 8.48 | 8.35 | |
| Qwen3-4BAccess Type=Open-source2025.12 | 8.4 | 8.42 | 7.84 | 8.31 | 8.31 | 8.18 | |
| Qwen3-8BAccess Type=Open-source2025.12 | 8.36 | 8.45 | 7.88 | 8.33 | 8.35 | 8.3 | |
| GPT-4.1Access Type=API-based2025.12 | 8.34 | 8.32 | 7.95 | 8.3 | 8.28 | 8.43 | |
| Qwen3-1.7BAccess Type=Open-source2025.12 | 8.3 | 8.35 | 7.86 | 8.23 | 8.15 | 7.95 | |
| Qwen2.5-7BAccess Type=Open-source2025.12 | 8.13 | 8.25 | 7.57 | 8.05 | 8.22 | 7.85 | |
| Llama-3.1-8BAccess Type=Open-source2025.12 | 8 | 8.24 | 7.46 | 7.89 | 8.09 | 7.8 | |
| Llama-3.2-3BAccess Type=Open-source2025.12 | 7.89 | 8.14 | 7.3 | 7.8 | 8.04 | 7.68 |