Creative Writing on Creative Writing v3 (Detailed Rubric Scoring)
83.3Overall Rubric ScoreHybrid Reward
Evaluation Results
| Method | Links | |||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Hybrid RewardBackbone=GLM-4.7-Flash2026.05 | 83.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Hybrid RewardBackbone=Qwen3-30B-A3B2026.05 | 82.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Qwen3-30B-A3B + Hybrid RLBackbone=Qwen3-30B-A3B, Training Method=Hybrid RL2026.05 | 82.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Rubicon-PreviewBackbone=Qwen3-30B-A3B, Training Method=rubric-anchor RL2026.05 | 82.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MiniMax-M2.5Category=Proprietary LLMs2026.06 | 80.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GLM-4.7-FlashRL Training=Baseline2026.05 | 80.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Hybrid RewardBackbone=Qwen3-4B2026.05 | 79.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Qwen3-30B-A3BRL Training=Baseline2026.05 | 77.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Qwen3-30B-A3BBackbone=Qwen3-30B-A3B2026.05 | 77.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Gemini-2.5-proCategory=Proprietary LLMs2026.06 | 77.19 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Qwen3-8B-OPERACategory=Our Methods, Technique=OPERA2026.06 | 72.89 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Qwen3-8B-SFTCategory=Our Methods2026.06 | 71.92 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Llama3.1-8B-OPERACategory=Our Methods, Technique=OPERA2026.06 | 67.09 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Qwen3-8BCategory=Our Methods2026.06 | 65.25 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| POPBackbone=Qwen-2.5-7B-Inst2026.04 | 64.71 | 1,577.32 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Llama3.1-8B-SFTCategory=Our Methods2026.06 | 63.21 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DeepWriter-8BCategory=Open-source LLMs2026.06 | 62.22 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GPT-4o-0513Category=Proprietary LLMs2026.06 | 61.76 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Qwen-2.5-7B-InstBackbone=Qwen-2.5-7B-Inst2026.04 | 59.96 | 1,524.83 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Qwen3-4BRL Training=Baseline2026.05 | 59.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| QwQ-32BStrategy=Rebalance2026.03 | 52.8 | 1,442.7 | 12.7 | 7.9 | 12.3 | 11.8 | 9.9 | 12.6 | 12.2 | 8.6 | 9.8 | 7.9 | 8 | 9.8 | 12.2 | 9.5 | 10.7 | |
| GPT3.5-TurboStrategy=Baseline2026.03 | 52.8 | 1,500 | 13.2 | 7.1 | 11.5 | 12.2 | 10.5 | 12.7 | 13.3 | 8.1 | 11 | 7.2 | 8.9 | 9.1 | 11.5 | 10.4 | 10.8 | |
| QwQ-32BStrategy=Baseline2026.03 | 52.7 | 1,438.2 | 12.6 | 8.1 | 12 | 11.7 | 9.8 | 13.6 | 12.6 | 8.5 | 9.8 | 8.1 | 8.3 | 9.8 | 11.4 | 9.6 | 10.6 | |
| LongWriter-Zero-32BCategory=Open-source LLMs2026.06 | 51.11 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Qwen3-14BStrategy=Rebalance2026.03 | 50.5 | 1,368 | 12.2 | 7.2 | 11.6 | 11.3 | 9.5 | 12.1 | 12.3 | 8.2 | 9.9 | 6.9 | 8 | 9.4 | 12.6 | 9 | 10.3 | |
| Hybrid RewardBackbone=DeepSeek-R1-Distill-Qwen-7B2026.05 | 44.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| POPBackbone=Qwen-2.5-7B2026.04 | 42.77 | 1,284.66 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Qwen3-14BStrategy=Baseline2026.03 | 42.6 | 1,295.2 | 8.8 | 7.1 | 10.3 | 7.7 | 8.2 | 9.2 | 11.2 | 6.4 | 8.1 | 5.5 | 6.6 | 8.1 | 10.8 | 8.4 | 8.8 | |
| Qwen-2.5-7BBackbone=Qwen-2.5-7B2026.04 | 41.18 | 1,200 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Train on DBackbone=Qwen-2.5-7B-Inst2026.04 | 40.38 | 1,026.52 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| AMARISRubric Strategy=per-instance rubric2026.05 | 40.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| AMARISRubric Strategy=global rubric2026.05 | 39.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RubricHub (RuRL)RL Training Approach=RuRL2026.05 | 39 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RuscaRLRL Training Approach=RuscaRL2026.05 | 38.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DeepSeek-R1-Distill-Qwen-7BRL Training=Baseline2026.05 | 37.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Llama3.1-8BCategory=Our Methods2026.06 | 37.79 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Naive (no RL)RL Training Approach=None2026.05 | 37.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LongWriter-8BCategory=Open-source LLMs2026.06 | 35.99 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Qwen2.5-7B-InstructStrategy=Baseline2026.03 | 33.5 | 877.6 | 7.4 | 5.2 | 7.2 | 7.6 | 8.5 | 6.1 | 8 | 6.2 | 8.7 | 4.8 | 6.9 | 4.9 | 9.5 | 10.6 | 6.3 | |
| Train on DBackbone=Qwen-2.5-7B2026.04 | 30.1 | 766.79 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DeepSeek-R1-Distill-Qwen-7BStrategy=Rebalance2026.03 | 23.4 | 349.7 | 4.3 | 4.5 | 4.7 | 5.1 | 6.7 | 3 | 5 | 4.6 | 6.6 | 4.1 | 5.4 | 2.9 | 9.9 | 9.9 | 3.5 | |
| DeepSeek-R1-Distill-Qwen-7BStrategy=Baseline2026.03 | 22.9 | 308.9 | 4 | 4.4 | 4.8 | 4.8 | 6.6 | 3.1 | 5 | 4.2 | 6.6 | 4 | 5.4 | 2.7 | 9.9 | 9.6 | 3.4 | |
| DeepSeek-R1-Distill-Qwen-1.5BStrategy=Rebalance2026.03 | 15.2 | 57.5 | 1.5 | 1.9 | 2.7 | 2.2 | 5.6 | 1.4 | 2.8 | 3 | 4.3 | 2.5 | 3.3 | 1.2 | 11.2 | 9.4 | 1.7 | |
| DeepSeek-R1-Distill-Qwen-1.5BStrategy=Baseline2026.03 | 15 | 30 | 1.3 | 1.8 | 2.8 | 1.9 | 5.4 | 1.5 | 2.9 | 2.9 | 4.1 | 2.5 | 3.2 | 1.2 | 11.5 | 9.2 | 1.8 |