Long-form Writing on LongBench-Write
95.42ScoreQwen-Plus
Evaluation Results
| Method | Links | |
|---|---|---|
| Qwen-PlusSFT=–, RL=–2025.06 | 95.42 | |
| Qwen2.5-14B-Writing-RLSFT=✓, RL=PPO2025.06 | 94.17 | |
| Qwen2.5-14B-WritingBench-SFTSFT=✓, RL=✗2025.06 | 93.75 | |
| Qwen2.5-7B-Writing-RLSFT=✓, RL=PPO2025.06 | 93.06 | |
| GPT-4oSFT=–, RL=–2025.06 | 92.92 | |
| Qwen2.5-7B-Reference-SFTSFT=✓, RL=✗2025.06 | 92.88 | |
| Qwen2.5-7B-WritingBench-SFT (24k)SFT=✓, RL=✗2025.06 | 92.65 | |
| Qwen2.5-7B-WritingBench-SFT (12k)SFT=✓, RL=✗2025.06 | 92.56 | |
| Llama3.1-8B-Writing-RLSFT=✓, RL=PPO2025.06 | 92.09 | |
| Qwen2.5-7B-Writing-RLSFT=✓, RL=GRPO2025.06 | 91.84 | |
| Llama3.1-8B-Reference-SFTSFT=✓, RL=✗2025.06 | 91.53 | |
| Llama3.1-8B-WritingBench-SFTSFT=✓, RL=✗2025.06 | 90.67 | |
| Qwen2.5-14B-InstructSFT=✗, RL=✗2025.06 | 89.2 | |
| Qwen2.5-7B-InstructSFT=✗, RL=✗2025.06 | 87.2 | |
| Longwriter-Zero-32BSFT=✓, RL=GRPO2025.06 | 85.9 | |
| Longwriter-9BSFT=✓, RL=DPO2025.06 | 80.83 | |
| Llama3.1-8B-InstructSFT=✗, RL=✗2025.06 | 80.79 | |
| Suri-7BSFT=✓, RL=✗2025.06 | 33.44 |