Math Reasoning on MATH 500 (Accuracy)
95AccuracyOurs
Evaluation Results
| Method | Links | |
|---|---|---|
| OursBackbone=Qwen3-MoE-30B-A3B, Sparsity=50%2026.06 | 95 | |
| OursQBackbone=Qwen3-MoE-30B-A3B, Sparsity=25%, Quantization=4-bit2026.06 | 94.5 | |
| BaselineBackbone=Qwen3-MoE-30B-A3B, Type=–2026.06 | 92.8 | |
| EGLR-OracleLModel=Qwen2.5-Math-7B-Instruct2026.06 | 88.2 | |
| EGLR-OracleLModel=Qwen2.5-7B-Instruct2026.06 | 88 | |
| EGLR-OracleLModel=Qwen2.5-14B-Instruct2026.06 | 86.2 | |
| EGLR-OracleLModel=Qwen2.5-Math-1.5B-Instruct2026.06 | 84.4 | |
| EGLR-SCL (B = 10)Model=Qwen2.5-Math-7B-Instruct2026.06 | 82.8 | |
| ORZModel=ORZ2025.04 | 81.4 | |
| EGLR-OracleLModel=Qwen2.5-3B-Instruct2026.06 | 81.2 | |
| EGLR (best L)Model=Qwen2.5-Math-7B-Instruct2026.06 | 81.2 | |
| GreedyModel=Qwen2.5-Math-7B-Instruct2026.06 | 81 | |
| EGLR-SCL (B = 10)Model=Qwen2.5-14B-Instruct2026.06 | 79.8 | |
| NEMOTRON-CROSSTHINKCategory=Single Source, Blend=Bonly_mr2025.04 | 78.6 | |
| NEMOTRON-CROSSTHINKCategory=Question Types, Blend=Bopen↑2025.04 | 78.4 | |
| NEMOTRON-CROSSTHINKCategory=Question Types, Blend=Bmcq↑2025.04 | 78 | |
| NEMOTRON-CROSSTHINKCategory=Data Usefulness, Blend=Bscore2025.04 | 78 | |
| NEMOTRON-CROSSTHINKCategory=Data Source, Blend=Bmr↑2025.04 | 77.8 | |
| EGLR-SCL (B = 10)Model=Qwen2.5-7B-Instruct2026.06 | 77.8 | |
| NEMOTRON-CROSSTHINKCategory=Data Source, Blend=Bgpr↑2025.04 | 77.6 | |
| EGLR (best L)Model=Qwen2.5-14B-Instruct2026.06 | 77.4 | |
| NEMOTRON-CROSSTHINKBlend=Bnd2025.04 | 77 | |
| GreedyModel=Qwen2.5-14B-Instruct2026.06 | 77 | |
| EGLR (best L)Model=Qwen2.5-7B-Instruct2026.06 | 75.2 | |
| EGLR-SCL (B = 10)Model=Qwen2.5-Math-1.5B-Instruct2026.06 | 74.4 | |
| GreedyModel=Qwen2.5-7B-Instruct2026.06 | 73.2 | |
| EGLR-OracleLModel=Llama-3.1-8B-Instruct2026.06 | 72.8 | |
| EGLR (best L)Model=Qwen2.5-Math-1.5B-Instruct2026.06 | 72.6 | |
| NEMOTRON-CROSSTHINKCategory=Single Source, Blend=Bonly_gpr2025.04 | 72.2 | |
| GreedyModel=Qwen2.5-Math-1.5B-Instruct2026.06 | 70.4 | |
| EGLR-SCL (B = 10)Model=Qwen2.5-3B-Instruct2026.06 | 69.2 | |
| EGLR (best L)Model=Qwen2.5-3B-Instruct2026.06 | 65 | |
| GreedyModel=Qwen2.5-3B-Instruct2026.06 | 64.6 | |
| EGLR-SCL (B = 10)Model=Llama-3.1-8B-Instruct2026.06 | 58 | |
| EGLR-OracleLModel=Qwen2.5-0.5B-Instruct2026.06 | 51.4 | |
| MModel=M2025.04 | 48.3 | |
| EGLR (best L)Model=Llama-3.1-8B-Instruct2026.06 | 47.4 | |
| GreedyModel=Llama-3.1-8B-Instruct2026.06 | 44.8 | |
| EGLR-OracleLModel=Mistral-7B-Instruct-v0.32026.06 | 36.2 | |
| EGLR-SCL (B = 10)Model=Qwen2.5-0.5B-Instruct2026.06 | 34.2 | |
| EGLR (best L)Model=Qwen2.5-0.5B-Instruct2026.06 | 25.6 | |
| GreedyModel=Qwen2.5-0.5B-Instruct2026.06 | 22 | |
| EGLR-SCL (B = 10)Model=Mistral-7B-Instruct-v0.32026.06 | 18.8 | |
| EGLR (best L)Model=Mistral-7B-Instruct-v0.32026.06 | 14.8 | |
| GreedyModel=Mistral-7B-Instruct-v0.32026.06 | 13.6 |