Creative Writing on ArenaHard Creative Writing v2.0
29WR ScoreDPWriter
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| DPWriterBackbone=Llama-3.2-3B-Instruct2026.01 | 29 | 22.56 | 15.45 | |
| GRPOBackbone=Llama-3.2-3B-Instruct2026.01 | 21.7 | 23.32 | 15.18 | |
| GAPOBackbone=Llama-3.2-3B-Instruct2026.01 | 20.5 | 23.57 | 15.37 | |
| DarlingBackbone=Llama-3.2-3B-Instruct2026.01 | 19.5 | 21.25 | 15.18 | |
| DPWriterBackbone=Qwen3-4B2026.01 | 13.9 | 23.65 | 17.68 | |
| GRPO-UnlikelinessBackbone=Qwen3-4B2026.01 | 12.1 | 22.79 | 15.92 | |
| GAPOBackbone=Qwen3-4B2026.01 | 11.8 | 23.16 | 16.27 | |
| GRPOBackbone=Qwen3-4B2026.01 | 11 | 22.27 | 15.82 | |
| DarlingBackbone=Qwen3-4B2026.01 | 10.1 | 21.67 | 16.21 | |
| Qwen3-4BBackbone=Qwen3-4B2026.01 | 9 | 17.33 | 12.09 | |
| Llama-3.2-3B-InstructBackbone=Llama-3.2-3B-Instruct2026.01 | 5.4 | 19.91 | 6.32 | |
| GRPO-UnlikelinessBackbone=Llama-3.2-3B-Instruct2026.01 | 2.5 | 24.87 | 6.17 | |
| Qwen3-4B-BaseBackbone=Qwen3-4B2026.01 | 1.9 | 33.27 | 6.26 |