Math Reasoning on AMC23 (Pass@1, #Tokens)
100Pass@1 AccuracyReBalance
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ReBalanceBackbone=Qwen3-14B2026.03 | 100 | 5,230 | |
| BaselineBackbone=Qwen3-14B2026.03 | 95 | 7,240 | |
| CoDBackbone=Qwen3-14B2026.03 | 95 | 5,360 | |
| DEERBackbone=Qwen3-14B2026.03 | 95 | 5,527 | |
| NoWaitBackbone=Qwen3-14B2026.03 | 95 | 5,050 | |
| Dynasor-CoTBackbone=Qwen3-14B2026.03 | 95 | 6,582 | |
| DEERBackbone=QwQ-32B2026.03 | 95 | 6,435 | |
| SEALBackbone=QwQ-32B2026.03 | 95 | 6,341 | |
| ReBalanceBackbone=QwQ-32B2026.03 | 95 | 6,064 | |
| CE-GPPOBase Model=DS-R1-Distill-Qwen-7B, Training Algorithm=CE-GPPO, Beta1 Hyperparameter=0.75, Beta2 Hyperparameter=12025.09 | 93.8 | — | |
| DAPOBase Model=DS-R1-Distill-Qwen-7B, Training Algorithm=DAPO2025.09 | 93.4 | — | |
| CE-GPPOBase Model=DS-R1-Distill-Qwen-7B, Training Algorithm=CE-GPPO, Beta1 Hyperparameter=0.5, Beta2 Hyperparameter=12025.09 | 92.8 | — | |
| ReBalanceBackbone=DeepSeek-R1-Distill-Qwen-7B2026.03 | 92.5 | 4,667 | |
| CoDBackbone=QwQ-32B2026.03 | 92.5 | 6,217 | |
| NoWaitBackbone=QwQ-32B2026.03 | 92.5 | 4,717 | |
| FlashThinkBackbone=QwQ-32B2026.03 | 92.5 | 6,702 | |
| DS-R1-Distill-Qwen-7BBase Model=DS-R1-Distill-Qwen-7B2025.09 | 90.6 | — | |
| SEALBackbone=Qwen3-14B2026.03 | 90 | 6,126 | |
| NoThinkingBackbone=QwQ-32B2026.03 | 90 | 7,119 | |
| Dynasor-CoTBackbone=QwQ-32B2026.03 | 90 | 6,544 | |
| TrimRBackbone=QwQ-32B2026.03 | 90 | 6,055 | |
| GRPOBase Model=DS-R1-Distill-Qwen-7B, Training Algorithm=GRPO2025.09 | 88.8 | — | |
| Manifold SteeringBackbone=DeepSeek-R1-Distill-Qwen-7B2026.03 | 87.5 | 4,440 | |
| NoThinkingBackbone=Qwen3-14B2026.03 | 87.5 | 4,503 | |
| BaselineBackbone=QwQ-32B2026.03 | 87.5 | 7,021 | |
| CE-GPPOBase Model=DS-R1-Distill-Qwen-1.5B, Training Algorithm=CE-GPPO, Beta1 Hyperparameter=0.5, Beta2 Hyperparameter=12025.09 | 85.9 | — | |
| CE-GPPOBase Model=DS-R1-Distill-Qwen-1.5B, Training Algorithm=CE-GPPO, Beta1 Hyperparameter=0.75, Beta2 Hyperparameter=12025.09 | 85.6 | — | |
| BaselineBackbone=DeepSeek-R1-Distill-Qwen-7B2026.03 | 85 | 6,375 | |
| CoDBackbone=DeepSeek-R1-Distill-Qwen-7B2026.03 | 85 | 3,654 | |
| NoWaitBackbone=DeepSeek-R1-Distill-Qwen-7B2026.03 | 85 | 3,795 | |
| Dynasor-CoTBackbone=DeepSeek-R1-Distill-Qwen-7B2026.03 | 85 | 5,121 | |
| DAPOBase Model=DS-R1-Distill-Qwen-1.5B, Training Algorithm=DAPO2025.09 | 84.4 | — | |
| MCPO-DAPOBackbone=Qwen3-8B2026.05 | 84.13 | — | |
| MCPO-DAPOBackbone=Qwen3-4B2026.05 | 83.44 | — | |
| DAPOBackbone=Qwen3-8B2026.05 | 82.97 | — | |
| MCPO-GRPOBackbone=Qwen3-8B2026.05 | 82.89 | — | |
| StaticBudget=40962026.04 | 82.5 | 11,269 | |
| MCPO-GRPOBackbone=Qwen3-4B2026.05 | 81.25 | — | |
| MT-GRPOBackbone=Qwen3-8B2026.05 | 80.94 | — | |
| DAPOBackbone=Qwen3-4B2026.05 | 80.83 | — | |
| TABModel Variant=4B, Budget=10k2026.04 | 80.5 | 6,916 | |
| MGSBackbone=Qwen3-8B2026.05 | 80.22 | — | |
| GRPOBackbone=Qwen3-8B2026.05 | 80.13 | — | |
| DEERBackbone=DeepSeek-R1-Distill-Qwen-7B2026.03 | 80 | 5,157 | |
| TABBudget=10k2026.04 | 80 | 8,212 | |
| TABModel Variant=4B, Budget=8k2026.04 | 79.8 | 5,846 | |
| TAB All-SubQBudget=10k2026.04 | 79.5 | 6,945 | |
| CLIPOBackbone=Qwen3-8B2026.05 | 79.38 | — | |
| GRPOBase Model=DS-R1-Distill-Qwen-1.5B, Training Algorithm=GRPO2025.09 | 79.3 | — | |
| GRPOBackbone=Qwen3-4B2026.05 | 78.81 | — | |
| MT-GRPOBackbone=Qwen3-4B2026.05 | 78.53 | — | |
| MGSBackbone=Qwen3-4B2026.05 | 78.1 | — | |
| CLIPOBackbone=Qwen3-4B2026.05 | 78.08 | — | |
| SEALBackbone=DeepSeek-R1-Distill-Qwen-7B2026.03 | 77.5 | 5,164 | |
| TABBudget=8k2026.04 | 77.5 | 5,587 | |
| TAB All-SubQBudget=8k2026.04 | 77.5 | 6,018 | |
| Base ModelBackbone=Qwen3-8B2026.05 | 76.25 | — | |
| SEALBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.03 | 75 | 5,084 | |
| DS-R1-Distill-Qwen-1.5BBase Model=DS-R1-Distill-Qwen-1.5B2025.09 | 73.7 | — | |
| TABModel Variant=4B, Budget=5k2026.04 | 73 | 3,731 | |
| Dynasor-CoTBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.03 | 72.5 | 6,505 | |
| Manifold SteeringBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.03 | 72.5 | 6,236 | |
| ReBalanceBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.03 | 72.5 | 5,744 | |
| StaticBudget=20482026.04 | 72.5 | 6,110 | |
| TABBudget=5k2026.04 | 72.5 | 3,392 | |
| TAB All-SubQBudget=5k2026.04 | 72.5 | 3,485 | |
| Base ModelBackbone=Qwen3-4B2026.05 | 71.17 | — | |
| L-JProtocol=Individual, Model Variant=4B2026.04 | 71 | 4,660 | |
| NEMOTRON-CROSSTHINKCategory=Single Source, Blend=Bonly_mr2025.04 | 70 | — | |
| L-JProtocol=Multi-Turn2026.04 | 70 | 5,923 | |
| L-JProtocol=Multi-Turn, Strategy=All-SubQ2026.04 | 70 | 5,882 | |
| L-JProtocol=Multi-Turn, Model Variant=4B2026.04 | 70 | 4,675 | |
| NEMOTRON-CROSSTHINKCategory=Data Source, Blend=Bmr↑2025.04 | 67.5 | — | |
| L-JProtocol=Individual2026.04 | 67.5 | 6,054 | |
| CGDTeacher Model=Claude Sonnet 3.72025.05 | 67.5 | — | |
| NoThinkingBackbone=DeepSeek-R1-Distill-Qwen-7B2026.03 | 65 | 1,911 | |
| NEMOTRON-CROSSTHINKCategory=Data Source, Blend=Bgpr↑2025.04 | 65 | — | |
| BaselineBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.03 | 62.5 | 8,239 | |
| CoDBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.03 | 62.5 | 6,812 | |
| ORZModel=ORZ2025.04 | 62.5 | — | |
| NEMOTRON-CROSSTHINKCategory=Question Types, Blend=Bopen↑2025.04 | 62.5 | — | |
| NEMOTRON-CROSSTHINKCategory=Data Usefulness, Blend=Bscore2025.04 | 62.5 | — | |
| StaticBudget=10242026.04 | 62.5 | 4,053 | |
| TABModel Variant=4B, Budget=3k2026.04 | 62.5 | 2,561 | |
| CFTTeacher Model=GPT-4o2025.05 | 62.5 | — | |
| CGDTeacher Model=S1.1-32B2025.05 | 62.5 | — | |
| TAB All-SubQBudget=3k2026.04 | 61 | 2,603 | |
| TABBudget=3k2026.04 | 60.5 | 2,436 | |
| NoThinkingBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.03 | 60 | 3,267 | |
| NoWaitBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.03 | 60 | 3,302 | |
| NEMOTRON-CROSSTHINKBlend=Bnd2025.04 | 60 | — | |
| NEMOTRON-CROSSTHINKCategory=Question Types, Blend=Bmcq↑2025.04 | 60 | — | |
| StaticBudget=5122026.04 | 60 | 2,708 | |
| DEERBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.03 | 57.5 | 5,132 | |
| StaticBudget=2562026.04 | 57.5 | 2,209 | |
| NEMOTRON-CROSSTHINKCategory=Single Source, Blend=Bonly_gpr2025.04 | 55 | — | |
| CFTTeacher Model=S1.1-32B2025.05 | 55 | — | |
| MModel=M2025.04 | 40 | — | |
| Qwen2.5-Math-7B (Base)2025.05 | 40 | — |