Human-centric Quality Evaluation on Arena-Hard
28.8Arena-Hard ScoreInstruct
Evaluation Results
| Method | Links | |
|---|---|---|
| InstructModel=Qwen3-4B2026.05 | 28.8 | |
| REDIPOModel=Qwen3-4B2026.05 | 28.5 | |
| DPOModel=Qwen3-4B2026.05 | 28.2 | |
| DivPOModel=Qwen3-4B2026.05 | 28.2 | |
| BaseModel=Qwen3-4B2026.05 | 19.5 | |
| InstructModel=LLaMA-3.1-8B2026.05 | 19.3 | |
| REDIPOModel=LLaMA-3.1-8B2026.05 | 18.6 | |
| DPOModel=LLaMA-3.1-8B2026.05 | 18.3 | |
| DivPOModel=LLaMA-3.1-8B2026.05 | 18.1 | |
| BaseModel=OLMo-3-7B2026.05 | 14.6 | |
| InstructModel=OLMo-3-7B2026.05 | 14 | |
| DPOModel=OLMo-3-7B2026.05 | 14 | |
| REDIPOModel=OLMo-3-7B2026.05 | 14 | |
| BaseModel=LLaMA-3.1-8B2026.05 | 13.4 | |
| DivPOModel=OLMo-3-7B2026.05 | 12.8 |