Instruction Following on IFEval (genuine-followup rate)
87.5Genuine-Followup RateQwen3-30B-A3B + Hybrid RL
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Qwen3-30B-A3B + Hybrid RLBackbone=Qwen3-30B-A3B, Training Method=Hybrid RL2026.05 | 87.5 | — | |
| Qwen3-30B-A3BBackbone=Qwen3-30B-A3B2026.05 | 85.4 | — | |
| Rubicon-PreviewBackbone=Qwen3-30B-A3B, Training Method=rubric-anchor RL2026.05 | 82.7 | — | |
| SCASModel=Qwen2.5-7B2026.05 | 68.58 | — | |
| GRACEModel=Qwen2.5-7B2026.05 | 62.7 | — | |
| PPLModel=Qwen2.5-7B2026.05 | 62.29 | — | |
| InfluenceModel=Qwen2.5-7B2026.05 | 62.29 | — | |
| IFDModel=Qwen2.5-7B2026.05 | 61.92 | — | |
| SCASModel=Qwen2.5-3B2026.05 | 61.37 | — | |
| RSR-ItemModel=Qwen2.5-7B2026.05 | 61 | — | |
| IFDModel=Qwen2.5-3B2026.05 | 60.48 | — | |
| PPLModel=Qwen2.5-3B2026.05 | 60.4 | — | |
| RouterModel=Qwen2.5-7B2026.05 | 60.26 | — | |
| RSRModel=Qwen2.5-7B2026.05 | 60.07 | — | |
| RouterModel=Qwen2.5-3B2026.05 | 59.89 | — | |
| InfluenceModel=Qwen2.5-3B2026.05 | 59.89 | — | |
| RSR-ItemModel=Qwen2.5-3B2026.05 | 59.15 | — | |
| GRACEModel=Qwen2.5-3B2026.05 | 58.75 | — | |
| RSRModel=Qwen2.5-3B2026.05 | 56.56 | — | |
| Precise FusionExpert Combination=Code + Instruction2026.04 | 53.97 | — | |
| Precise FusionExpert Combination=Code + Instruction + Math2026.04 | 42.14 | — | |
| Avg BaselineExpert Combination=Code + Instruction2026.04 | 40.3 | — | |
| Precise FusionExpert Combination=Instruction + Math2026.04 | 37.87 | — | |
| DARE TiesExpert Combination=Code + Instruction + Math2026.04 | 35.75 | — | |
| Instruction TunedModel Category=Base Models2026.04 | 35.51 | — | |
| Ties MergingExpert Combination=Instruction + Math2026.04 | 34.04 | — | |
| DARE TiesExpert Combination=Instruction + Math2026.04 | 33.89 | — | |
| DARE Task ArithmeticExpert Combination=Code + Instruction2026.04 | 33.42 | — | |
| DARE TiesExpert Combination=Code + Instruction2026.04 | 33.28 | — | |
| DARE Task ArithmeticExpert Combination=Code + Instruction + Math2026.04 | 33.2 | — | |
| Task ArithmeticExpert Combination=Code + Instruction + Math2026.04 | 32.87 | — | |
| DARE Task ArithmeticExpert Combination=Instruction + Math2026.04 | 32.35 | — | |
| Task ArithmeticExpert Combination=Code + Instruction2026.04 | 31.6 | — | |
| WIDENExpert Combination=Code + Instruction + Math2026.04 | 31.44 | — | |
| Task ArithmeticExpert Combination=Instruction + Math2026.04 | 31.31 | — | |
| WIDENExpert Combination=Instruction + Math2026.04 | 30.82 | — | |
| WIDENExpert Combination=Code + Instruction2026.04 | 30.42 | — | |
| Avg BaselineExpert Combination=Instruction + Math2026.04 | 29.11 | — | |
| Avg BaselineExpert Combination=Code + Instruction + Math2026.04 | 27.37 | — | |
| Ties MergingExpert Combination=Code + Instruction + Math2026.04 | 26.95 | — | |
| Ties MergingExpert Combination=Code + Instruction2026.04 | 24.48 | — | |
| SCASModel=Qwen2.5-0.5B2026.05 | 24.4 | — | |
| Precise FusionExpert Combination=Code + Math2026.04 | 24.35 | — | |
| IFDModel=Qwen2.5-0.5B2026.05 | 24.03 | — | |
| WIDENExpert Combination=Code + Math2026.04 | 24.02 | — | |
| RouterModel=Qwen2.5-0.5B2026.05 | 23.84 | — | |
| BaseModel Category=Base Models2026.04 | 23.59 | — | |
| CodeModel Category=Base Models2026.04 | 23.55 | — | |
| RSR-ItemModel=Qwen2.5-0.5B2026.05 | 23.47 | — | |
| GRACEModel=Qwen2.5-0.5B2026.05 | 23.32 | — | |
| PPLModel=Qwen2.5-0.5B2026.05 | 23.29 | — | |
| InfluenceModel=Qwen2.5-0.5B2026.05 | 23.29 | — | |
| RSRModel=Qwen2.5-0.5B2026.05 | 22.77 | — | |
| DARE TiesExpert Combination=Code + Math2026.04 | 22.53 | — | |
| DARE Task ArithmeticExpert Combination=Code + Math2026.04 | 22.02 | — | |
| Task ArithmeticExpert Combination=Code + Math2026.04 | 21.92 | — | |
| MathModel Category=Base Models2026.04 | 21.8 | — | |
| Ties MergingExpert Combination=Code + Math2026.04 | 21.63 | — | |
| Avg BaselineExpert Combination=Code + Math2026.04 | 21.05 | — | |
| Qwen3.5-397B-A17B2026.04 | 10.5 | — | |
| Qwen3.5-9B2026.04 | 5.4 | — | |
| gpt-oss-120b2026.04 | 3.7 | — | |
| gpt-oss-20b2026.04 | 3.3 | — | |
| GLM-4.72026.04 | 3 | — | |
| Qwen3.5-27B2026.04 | 1.1 | — | |
| Claude-Opus-4.7Stage=-2026.05 | — | 89.1 | |
| Conifer-7B-DPOStage=-2026.05 | — | 52.3 | |
| Crab-7B-DPOStage=-2026.05 | — | 57.7 | |
| Distill-Qwen-14BStage=BASE2026.05 | — | 74.9 | |
| Distill-Qwen-14BStage=Iter12026.05 | — | 76.4 | |
| Distill-Qwen-14BStage=Iter22026.05 | — | 78 | |
| Distill-Qwen-14BStage=Iter32026.05 | — | 80 | |
| GPT-4oStage=-2026.05 | — | 84.8 | |
| Llama-3.1-8B-InstructStage=BASE2026.05 | — | 73.8 | |
| Llama-3.1-8B-InstructStage=Iter12026.05 | — | 76.9 | |
| Llama-3.1-8B-InstructStage=Iter22026.05 | — | 78 | |
| Llama-3.1-8B-InstructStage=Iter32026.05 | — | 78.4 | |
| Qwen2.5-1.5B-InstructStage=BASE2026.05 | — | 43.6 | |
| Qwen2.5-1.5B-InstructStage=Iter12026.05 | — | 44.7 | |
| Qwen2.5-1.5B-InstructStage=Iter22026.05 | — | 47.1 | |
| Qwen2.5-1.5B-InstructStage=Iter32026.05 | — | 47.5 | |
| Qwen2.5-7B-InstructStage=BASE2026.05 | — | 73.9 | |
| Qwen2.5-7B-InstructStage=Iter12026.05 | — | 76.5 | |
| Qwen2.5-7B-InstructStage=Iter22026.05 | — | 77.3 | |
| Qwen2.5-7B-InstructStage=Iter32026.05 | — | 78.6 | |
| QwQ-32BStage=-2026.05 | — | 83.9 | |
| R1-0528-Qwen3-8BStage=BASE2026.05 | — | 79.7 | |
| R1-0528-Qwen3-8BStage=Iter12026.05 | — | 81.3 | |
| R1-0528-Qwen3-8BStage=Iter22026.05 | — | 81.9 | |
| R1-0528-Qwen3-8BStage=Iter32026.05 | — | 81.9 | |
| RAIF-7BStage=-2026.05 | — | 74.1 | |
| Self-Supervised-7BStage=-2026.05 | — | 78.9 | |
| SPAR-8B-DPOStage=-2026.05 | — | 82.4 | |
| VERIF-8BStage=-2026.05 | — | 87.1 |