General Chat on WildBench
68.16LLM Judge ScoreGOLF
Evaluation results
| Method | Links | |
|---|---|---|
| GOLFModel Backbone=Qwen-3-8B2026.03 | 68.16 | |
| Pairwise-GRPOModel Backbone=Qwen-3-8B2026.03 | 67.77 | |
| Rubric-as-RewardModel Backbone=Qwen-3-8B2026.03 | 67.09 | |
| Critique-GRPOModel Backbone=Qwen-3-8B2026.03 | 64.84 | |
| Direct-LikertModel Backbone=Qwen-3-8B2026.03 | 58.01 | |
| Qwen-3-8BModel Backbone=Qwen-3-8B2026.03 | 48.05 | |
| GOLFModel Backbone=Llama-3.1-8B-Instruct2026.03 | 34.42 | |
| Rubric-as-RewardModel Backbone=Llama-3.1-8B-Instruct2026.03 | 26.51 | |
| Pairwise-GRPOModel Backbone=Llama-3.1-8B-Instruct2026.03 | 25.54 | |
| Critique-GRPOModel Backbone=Llama-3.1-8B-Instruct2026.03 | 25.09 | |
| Direct-LikertModel Backbone=Llama-3.1-8B-Instruct2026.03 | 13.48 | |
| Llama-3.1-8B-InstructModel Backbone=Llama-3.1-8B-Instruct2026.03 | -8.25 |