Multi-turn instruction following on SysBench
92.31CSRImpRIF-32B
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| ImpRIF-32BTraining Stage=SFT+RL2026.02 | 92.31 | 86.04 | 69.2 | |
| ImpRIF-32BTraining Stage=SFT2026.02 | 91.3 | 83.6 | 65.4 | |
| ImpRIF-32BTraining Stage=RL2026.02 | 88.46 | 78.8 | 57.6 | |
| ImpRIF-8BTraining Stage=SFT+RL2026.02 | 88.41 | 79.08 | 59.6 | |
| Qwen3-235B-A22BTraining Stage=Base2026.02 | 87.75 | 76.84 | 55.6 | |
| Qwen3-14BTraining Stage=Base2026.02 | 86.68 | 75.08 | 53.2 | |
| ImpRIF-8BTraining Stage=SFT2026.02 | 86.24 | 75.64 | 53.76 | |
| Qwen3-32BTraining Stage=Base2026.02 | 86.19 | 73.68 | 50.4 | |
| ImpRIF-8BTraining Stage=RL2026.02 | 84.66 | 71.48 | 47.2 | |
| ImpRIF-4BTraining Stage=SFT+RL2026.02 | 82.75 | 70.68 | 46 | |
| Qwen3-8BTraining Stage=Base2026.02 | 81.98 | 66.52 | 42 | |
| ImpRIF-4BTraining Stage=SFT2026.02 | 81.82 | 67.08 | 41.64 | |
| Ministral3-14BTraining Stage=Base2026.02 | 81.22 | 66.84 | 44 | |
| Qwen2.5-72BReasoning Capability=Lacks inference reasoning capabilities2026.02 | 80.12 | 63.8 | 42.4 | |
| Qwen3-4BTraining Stage=Base2026.02 | 77.72 | 60.2 | 35.2 | |
| ImpRIF-4BTraining Stage=RL2026.02 | 77.1 | 61.2 | 37.8 | |
| Llama3.1-70BReasoning Capability=Lacks inference reasoning capabilities2026.02 | 75.87 | 58.48 | 36.4 | |
| AGD_LRPeModel=Llama 3.1 (8B)2025.09 | 74.3 | 58.2 | 33.9 | |
| AGD_LRPModel=Qwen 2.5 (7B)2025.09 | 73.7 | 56.4 | 32.7 | |
| AGD_LRPModel=Llama 3.1 (8B)2025.09 | 73.3 | 57.3 | 32.2 | |
| AGD_LRPModel=Gemma 3 (4B)2025.09 | 73.2 | 57.8 | 36.5 | |
| CADModel=Gemma 3 (4B)2025.09 | 73 | 57.9 | 36 | |
| AGD_LRPeModel=Gemma 3 (4B)2025.09 | 73 | 57.9 | 36 | |
| CADModel=Llama 3.1 (8B)2025.09 | 72.2 | 58.8 | 32.3 | |
| AGD_LRPeModel=Qwen 2.5 (7B)2025.09 | 71.1 | 53 | 29.9 | |
| GreedyModel=Gemma 3 (4B)2025.09 | 69.8 | 52.4 | 33.3 | |
| NucleusModel=Gemma 3 (4B)2025.09 | 69.3 | 52.2 | 33.2 | |
| AGD_I×GeModel=Gemma 3 (4B)2025.09 | 69 | 51.6 | 32.2 | |
| AGD_I×GModel=Gemma 3 (4B)2025.09 | 68.5 | 50.4 | 31.8 | |
| AGD_I×GeModel=Llama 3.1 (8B)2025.09 | 67.8 | 50.1 | 27.2 | |
| GreedyModel=Qwen 2.5 (7B)2025.09 | 67.6 | 47.9 | 27.1 | |
| AGD_I×GeModel=Qwen 2.5 (7B)2025.09 | 67.3 | 46.9 | 25.1 | |
| GreedyModel=Llama 3.1 (8B)2025.09 | 67.1 | 48.4 | 26 | |
| AGD_I×GModel=Qwen 2.5 (7B)2025.09 | 66.8 | 46.4 | 25.2 | |
| CADModel=Qwen 2.5 (7B)2025.09 | 65.7 | 49.2 | 25.2 | |
| AGD_I×GModel=Llama 3.1 (8B)2025.09 | 65.1 | 46.5 | 24.2 | |
| NucleusModel=Qwen 2.5 (7B)2025.09 | 64.8 | 44.8 | 24.7 | |
| NucleusModel=Llama 3.1 (8B)2025.09 | 58 | 40.6 | 20.2 |