Long-form writing on LongBench-Write (test)
86.7Overall Quality Score (S_bar)Claude Opus 4.6
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Claude Opus 4.62026.06 | 86.7 | 88.1 | 98.4 | |
| LongWriter-Zero-32BParameters=32B2025.06 | 86.3 | 84.2 | 88.4 | |
| LongWriter-9B-DPOParameters=9B, Optimization=DPO2025.06 | 84 | 82.6 | 85.4 | |
| GPT-5.42026.06 | 83.3 | 84.7 | 98.4 | |
| POLARISBackbone=Qwen3.5-9B2026.06 | 81.2 | 88.5 | 90.2 | |
| Claude 3.5 SonnetModel Version=3.5 Sonnet2025.06 | 80.7 | 73.7 | 87.7 | |
| LongWriter-9BParameters=9B2025.06 | 80.5 | 78.6 | 82.3 | |
| Gemini 3 Flash2026.06 | 80.3 | 83.2 | 96.8 | |
| LongWriter-8BParameters=8B2025.06 | 79.8 | 77.4 | 82.2 | |
| Gemini 3.1 Pro2026.06 | 79.8 | 84.8 | 94.2 | |
| Gemma-4-E4BPLE=true2026.06 | 78.4 | 82.7 | 94.9 | |
| Qwen3.5-27B2026.06 | 77.8 | 91.5 | 85.3 | |
| Gemma-4-31B2026.06 | 77.6 | 81.1 | 95.8 | |
| Mistral-Large-InstructType=Instruct2025.06 | 77 | 65.6 | 88.3 | |
| Gemma-4-26B-A4BMoE=true2026.06 | 73.4 | 75.5 | 97.6 | |
| GLM-4-9B-ChatParameters=9B, Type=Chat2025.06 | 68.3 | 51 | 85.5 | |
| GPT-4 TurboModel Version=Turbo2025.06 | 67.3 | 47.9 | 86.6 | |
| Qwen3.5-9B2026.06 | 67.1 | 90.3 | 75.1 | |
| Llama-3.1-70B-InstructParameters=70B, Type=Instruct2025.06 | 65.6 | 50.8 | 80.3 | |
| Llama-3.1-8B-InstructParameters=8B, Type=Instruct2025.06 | 60.3 | 50 | 70.6 | |
| LongWriter-Zero-32B2026.06 | 58.3 | 73 | 76.8 | |
| LongWriter-Llama3.1-8B2026.06 | 57.8 | 75.9 | 71.4 | |
| Ministral-3-8B2026.06 | 57.5 | 67.9 | 80.5 | |
| Suri-I-ORPOOptimization=ORPO2025.06 | 56.6 | 59.6 | 53.5 | |
| DeepSeek-R1-Distill-Qwen-14B2026.06 | 41.5 | 53.6 | 71.9 |