List your company

Math Reasoning on AMC23 (Pass@1, #Tokens)

100Pass@1 AccuracyReBalance
37.653.87086.2Apr 15, 2025Jun 21, 2025Aug 28, 2025Nov 3, 2025Jan 10, 2026Mar 18, 2026May 25, 2026

Evaluation results

MethodLinks
ReBalanceBackbone=Qwen3-14B2026.031005,230
BaselineBackbone=Qwen3-14B2026.03957,240
CoDBackbone=Qwen3-14B2026.03955,360
DEERBackbone=Qwen3-14B2026.03955,527
NoWaitBackbone=Qwen3-14B2026.03955,050
Dynasor-CoTBackbone=Qwen3-14B2026.03956,582
DEERBackbone=QwQ-32B2026.03956,435
SEALBackbone=QwQ-32B2026.03956,341
ReBalanceBackbone=QwQ-32B2026.03956,064
CE-GPPOBase Model=DS-R1-Distill-Qwen-7B, Training Algorithm=CE-GPPO, Beta1 Hyperparameter=0.75, Beta2 Hyperparameter=12025.0993.8—
DAPOBase Model=DS-R1-Distill-Qwen-7B, Training Algorithm=DAPO2025.0993.4—
CE-GPPOBase Model=DS-R1-Distill-Qwen-7B, Training Algorithm=CE-GPPO, Beta1 Hyperparameter=0.5, Beta2 Hyperparameter=12025.0992.8—
ReBalanceBackbone=DeepSeek-R1-Distill-Qwen-7B2026.0392.54,667
CoDBackbone=QwQ-32B2026.0392.56,217
NoWaitBackbone=QwQ-32B2026.0392.54,717
FlashThinkBackbone=QwQ-32B2026.0392.56,702
DS-R1-Distill-Qwen-7BBase Model=DS-R1-Distill-Qwen-7B2025.0990.6—
SEALBackbone=Qwen3-14B2026.03906,126
NoThinkingBackbone=QwQ-32B2026.03907,119
Dynasor-CoTBackbone=QwQ-32B2026.03906,544
TrimRBackbone=QwQ-32B2026.03906,055
GRPOBase Model=DS-R1-Distill-Qwen-7B, Training Algorithm=GRPO2025.0988.8—
Manifold SteeringBackbone=DeepSeek-R1-Distill-Qwen-7B2026.0387.54,440
NoThinkingBackbone=Qwen3-14B2026.0387.54,503
BaselineBackbone=QwQ-32B2026.0387.57,021
CE-GPPOBase Model=DS-R1-Distill-Qwen-1.5B, Training Algorithm=CE-GPPO, Beta1 Hyperparameter=0.5, Beta2 Hyperparameter=12025.0985.9—
CE-GPPOBase Model=DS-R1-Distill-Qwen-1.5B, Training Algorithm=CE-GPPO, Beta1 Hyperparameter=0.75, Beta2 Hyperparameter=12025.0985.6—
BaselineBackbone=DeepSeek-R1-Distill-Qwen-7B2026.03856,375
CoDBackbone=DeepSeek-R1-Distill-Qwen-7B2026.03853,654
NoWaitBackbone=DeepSeek-R1-Distill-Qwen-7B2026.03853,795
Dynasor-CoTBackbone=DeepSeek-R1-Distill-Qwen-7B2026.03855,121
DAPOBase Model=DS-R1-Distill-Qwen-1.5B, Training Algorithm=DAPO2025.0984.4—
MCPO-DAPOBackbone=Qwen3-8B2026.0584.13—
MCPO-DAPOBackbone=Qwen3-4B2026.0583.44—
DAPOBackbone=Qwen3-8B2026.0582.97—
MCPO-GRPOBackbone=Qwen3-8B2026.0582.89—
StaticBudget=40962026.0482.511,269
MCPO-GRPOBackbone=Qwen3-4B2026.0581.25—
MT-GRPOBackbone=Qwen3-8B2026.0580.94—
DAPOBackbone=Qwen3-4B2026.0580.83—
TABModel Variant=4B, Budget=10k2026.0480.56,916
MGSBackbone=Qwen3-8B2026.0580.22—
GRPOBackbone=Qwen3-8B2026.0580.13—
DEERBackbone=DeepSeek-R1-Distill-Qwen-7B2026.03805,157
TABBudget=10k2026.04808,212
TABModel Variant=4B, Budget=8k2026.0479.85,846
TAB All-SubQBudget=10k2026.0479.56,945
CLIPOBackbone=Qwen3-8B2026.0579.38—
GRPOBase Model=DS-R1-Distill-Qwen-1.5B, Training Algorithm=GRPO2025.0979.3—
GRPOBackbone=Qwen3-4B2026.0578.81—
MT-GRPOBackbone=Qwen3-4B2026.0578.53—
MGSBackbone=Qwen3-4B2026.0578.1—
CLIPOBackbone=Qwen3-4B2026.0578.08—
SEALBackbone=DeepSeek-R1-Distill-Qwen-7B2026.0377.55,164
TABBudget=8k2026.0477.55,587
TAB All-SubQBudget=8k2026.0477.56,018
Base ModelBackbone=Qwen3-8B2026.0576.25—
SEALBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.03755,084
DS-R1-Distill-Qwen-1.5BBase Model=DS-R1-Distill-Qwen-1.5B2025.0973.7—
TABModel Variant=4B, Budget=5k2026.04733,731
Dynasor-CoTBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.0372.56,505
Manifold SteeringBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.0372.56,236
ReBalanceBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.0372.55,744
StaticBudget=20482026.0472.56,110
TABBudget=5k2026.0472.53,392
TAB All-SubQBudget=5k2026.0472.53,485
Base ModelBackbone=Qwen3-4B2026.0571.17—
L-JProtocol=Individual, Model Variant=4B2026.04714,660
NEMOTRON-CROSSTHINKCategory=Single Source, Blend=Bonly_mr2025.0470—
L-JProtocol=Multi-Turn2026.04705,923
L-JProtocol=Multi-Turn, Strategy=All-SubQ2026.04705,882
L-JProtocol=Multi-Turn, Model Variant=4B2026.04704,675
NEMOTRON-CROSSTHINKCategory=Data Source, Blend=Bmr↑2025.0467.5—
L-JProtocol=Individual2026.0467.56,054
CGDTeacher Model=Claude Sonnet 3.72025.0567.5—
NoThinkingBackbone=DeepSeek-R1-Distill-Qwen-7B2026.03651,911
NEMOTRON-CROSSTHINKCategory=Data Source, Blend=Bgpr↑2025.0465—
BaselineBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.0362.58,239
CoDBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.0362.56,812
ORZModel=ORZ2025.0462.5—
NEMOTRON-CROSSTHINKCategory=Question Types, Blend=Bopen↑2025.0462.5—
NEMOTRON-CROSSTHINKCategory=Data Usefulness, Blend=Bscore2025.0462.5—
StaticBudget=10242026.0462.54,053
TABModel Variant=4B, Budget=3k2026.0462.52,561
CFTTeacher Model=GPT-4o2025.0562.5—
CGDTeacher Model=S1.1-32B2025.0562.5—
TAB All-SubQBudget=3k2026.04612,603
TABBudget=3k2026.0460.52,436
NoThinkingBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.03603,267
NoWaitBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.03603,302
NEMOTRON-CROSSTHINKBlend=Bnd2025.0460—
NEMOTRON-CROSSTHINKCategory=Question Types, Blend=Bmcq↑2025.0460—
StaticBudget=5122026.04602,708
DEERBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.0357.55,132
StaticBudget=2562026.0457.52,209
NEMOTRON-CROSSTHINKCategory=Single Source, Blend=Bonly_gpr2025.0455—
CFTTeacher Model=S1.1-32B2025.0555—
MModel=M2025.0440—
Qwen2.5-Math-7B (Base)2025.0540—