Math Reasoning on GSM8K (Pass@1 & Token Count)
96.8Pass@1 AccuracyReBalance
Evaluation results
| Method | Links | ||
|---|---|---|---|
| ReBalanceBackbone=QwQ-32B2026.03 | 96.8 | 1,289 | |
| NoThinkingBackbone=QwQ-32B2026.03 | 96.5 | 1,326 | |
| FlashThinkBackbone=QwQ-32B2026.03 | 96.5 | 910 | |
| ReBalanceBackbone=Qwen3-14B2026.03 | 96.3 | 1,441 | |
| BaselineBackbone=QwQ-32B2026.03 | 96.3 | 1,506 | |
| NoWaitBackbone=QwQ-32B2026.03 | 96.3 | 942 | |
| CoDBackbone=QwQ-32B2026.03 | 96.2 | 670 | |
| DEERBackbone=QwQ-32B2026.03 | 96.2 | 944 | |
| SEALBackbone=QwQ-32B2026.03 | 96.2 | 1,221 | |
| DEERBackbone=Qwen3-14B2026.03 | 95.8 | 934 | |
| SEALBackbone=Qwen3-14B2026.03 | 95.7 | 1,369 | |
| CoDBackbone=Qwen3-14B2026.03 | 95.6 | 627 | |
| NoWaitBackbone=Qwen3-14B2026.03 | 95.6 | 1,129 | |
| Dynasor-CoTBackbone=Qwen3-14B2026.03 | 95.6 | 1,483 | |
| Dynasor-CoTBackbone=QwQ-32B2026.03 | 95.2 | 1,095 | |
| BaselineBackbone=Qwen3-14B2026.03 | 95.1 | 2,231 | |
| NoThinkingBackbone=Qwen3-14B2026.03 | 95.1 | 369 | |
| TrimRBackbone=QwQ-32B2026.03 | 93.7 | 1,319 | |
| ReBalanceBackbone=DeepSeek-R1-Distill-Qwen-7B2026.03 | 91.6 | 912 | |
| BaseModel=Qwen2.5-7B-Instruct2026.05 | 91.43 | — | |
| DEERBackbone=DeepSeek-R1-Distill-Qwen-7B2026.03 | 90.4 | 676 | |
| SimCTTeacher=Qwen2.5-7B-Instruct, Student=Phi-4-mini-instruct2026.05 | 90.22 | — | |
| NoWaitBackbone=DeepSeek-R1-Distill-Qwen-7B2026.03 | 90.2 | 806 | |
| BaselineBackbone=DeepSeek-R1-Distill-Qwen-7B2026.03 | 89.2 | 1,098 | |
| DSKDTeacher=Qwen2.5-7B-Instruct, Student=Phi-4-mini-instruct2026.05 | 88.63 | — | |
| SEALBackbone=DeepSeek-R1-Distill-Qwen-7B2026.03 | 88.4 | 811 | |
| GOLDTeacher=Qwen2.5-7B-Instruct, Student=Phi-4-mini-instruct2026.05 | 88.32 | — | |
| ALMTeacher=Qwen2.5-7B-Instruct, Student=Phi-4-mini-instruct2026.05 | 87.81 | — | |
| Dynasor-CoTBackbone=DeepSeek-R1-Distill-Qwen-7B2026.03 | 87.6 | 732 | |
| Manifold SteeringBackbone=DeepSeek-R1-Distill-Qwen-7B2026.03 | 87.6 | 440 | |
| NoThinkingBackbone=DeepSeek-R1-Distill-Qwen-7B2026.03 | 87.1 | 284 | |
| SimpleOPDTeacher=Qwen2.5-7B-Instruct, Student=Phi-4-mini-instruct2026.05 | 87.04 | — | |
| SFTTeacher=Qwen2.5-7B-Instruct, Student=Phi-4-mini-instruct2026.05 | 86.06 | — | |
| BaseModel=Phi-4-mini-instruct2026.05 | 85.75 | — | |
| CoDBackbone=DeepSeek-R1-Distill-Qwen-7B2026.03 | 84.5 | 339 | |
| ReBalanceBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.03 | 78.3 | 765 | |
| Manifold SteeringBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.03 | 77.2 | 593 | |
| Dynasor-CoTBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.03 | 77.1 | 1,035 | |
| SEALBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.03 | 76.4 | 754 | |
| BaselineBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.03 | 76 | 1,018 | |
| NoWaitBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.03 | 75.1 | 641 | |
| GRPO-MANumber of thoughts (TN)=4, Number of answers per thought (AN)=42025.09 | 70.58 | — | |
| CoDBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.03 | 69.5 | 531 | |
| DEERBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.03 | 69.2 | 684 | |
| GRPONumber of thoughts (TN)=16, Number of answers per thought (AN)=12025.09 | 65.13 | — | |
| GRPONumber of thoughts (TN)=8, Number of answers per thought (AN)=12025.09 | 64.67 | — | |
| SimCTTeacher=Qwen2.5-7B-Instruct, Student=Gemma-2-2B-it2026.05 | 63.53 | — | |
| SimCTTeacher=Phi-4-mini-instruct, Student=Gemma-2-2B-it2026.05 | 62.92 | — | |
| DSKDTeacher=Qwen2.5-7B-Instruct, Student=Gemma-2-2B-it2026.05 | 62.65 | — | |
| SimpleOPDTeacher=Phi-4-mini-instruct, Student=Gemma-2-2B-it2026.05 | 62.62 | — | |
| GRPONumber of thoughts (TN)=4, Number of answers per thought (AN)=12025.09 | 62.54 | — | |
| GOLDTeacher=Phi-4-mini-instruct, Student=Gemma-2-2B-it2026.05 | 62.52 | — | |
| ALMTeacher=Qwen2.5-7B-Instruct, Student=Gemma-2-2B-it2026.05 | 62.39 | — | |
| DSKDTeacher=Phi-4-mini-instruct, Student=Gemma-2-2B-it2026.05 | 62.37 | — | |
| SFTTeacher=Phi-4-mini-instruct, Student=Gemma-2-2B-it2026.05 | 62.32 | — | |
| ALMTeacher=Phi-4-mini-instruct, Student=Gemma-2-2B-it2026.05 | 62.24 | — | |
| GOLDTeacher=Qwen2.5-7B-Instruct, Student=Gemma-2-2B-it2026.05 | 62.22 | — | |
| SimpleOPDTeacher=Qwen2.5-7B-Instruct, Student=Gemma-2-2B-it2026.05 | 61.9 | — | |
| SFTTeacher=Qwen2.5-7B-Instruct, Student=Gemma-2-2B-it2026.05 | 61.86 | — | |
| BaseModel=Gemma-2-2B-it2026.05 | 61.85 | — | |
| NoThinkingBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.03 | 61.6 | 285 |