General Chat on ArenaHard v1.0
82.75Win RateDirect-Likert
Evaluation Results
| Method | Links | |
|---|---|---|
| Direct-LikertModel Backbone=Qwen-3-8B2026.03 | 82.75 | |
| Critique-GRPOModel Backbone=Qwen-3-8B2026.03 | 81.95 | |
| Rubric-as-RewardModel Backbone=Qwen-3-8B2026.03 | 81.9 | |
| Pairwise-GRPOModel Backbone=Qwen-3-8B2026.03 | 81.2 | |
| GOLFModel Backbone=Qwen-3-8B2026.03 | 80.9 | |
| Qwen-3-8BModel Backbone=Qwen-3-8B2026.03 | 70.7 | |
| GOLFModel Backbone=Llama-3.1-8B-Instruct2026.03 | 52.4 | |
| Rubric-as-RewardModel Backbone=Llama-3.1-8B-Instruct2026.03 | 52.1 | |
| Direct-LikertModel Backbone=Llama-3.1-8B-Instruct2026.03 | 51.55 | |
| Critique-GRPOModel Backbone=Llama-3.1-8B-Instruct2026.03 | 50.15 | |
| Pairwise-GRPOModel Backbone=Llama-3.1-8B-Instruct2026.03 | 49.2 | |
| Llama-3.1-8B-InstructModel Backbone=Llama-3.1-8B-Instruct2026.03 | 30.8 |