List your company
ResearchFormerly WizWand

Mathematical Reasoning on AMC 23 (Accuracy, Avg, Speedup)

95AccuracyGRPO
-1.223.77548.7573.725Mar 25, 2026Apr 6, 2026Apr 19, 2026May 1, 2026May 14, 2026May 26, 2026Jun 8, 2026
Updated 4mo ago

Evaluation results

MethodLinks
GRPOBackbone=Qwen-3-8B-Base, Method=GRPO2026.039556.66—
GRPO + ARROLBackbone=Qwen-3-8B-Base, Method=GRPO + ARROL2026.039559.531.62
GRPO + ARROLBackbone=Qwen-3-4B-Base, Method=GRPO + ARROL2026.0392.553.541.63
Skywork-OR1-Math-7BDistillation Domain=Single-Domain, Teacher Model=Skywork-OR1-Math-7B2026.0592.34——
Skywork-OR1-7BDistillation Domain=Multi-Domain, Teacher Model=Skywork-OR1-7B2026.0591.79——
GRPOBackbone=Qwen-3-4B-Base, Method=GRPO2026.0387.551.01—
GRPO + ARROLBackbone=Qwen-3-1.7B-Base, Method=GRPO + ARROL2026.0382.537.091.61
TrOPDDistillation Domain=Single-Domain, Teacher Model=Skywork-OR1-Math-7B, Student Model=DeepSeek-R1-Distill-Qwen-1.5B, Training Steps=200, Learning Rate=5 x 10^-6, Sample Rollouts=4, Max Generation Length=80962026.0577.03——
REOPOLDDistillation Domain=Single-Domain, Teacher Model=Skywork-OR1-Math-7B, Student Model=DeepSeek-R1-Distill-Qwen-1.5B, Training Steps=200, Learning Rate=5 x 10^-6, Sample Rollouts=4, Max Generation Length=80962026.0575.78——
OPDDistillation Domain=Single-Domain, Teacher Model=Skywork-OR1-Math-7B, Student Model=DeepSeek-R1-Distill-Qwen-1.5B, Training Steps=200, Learning Rate=5 x 10^-6, Sample Rollouts=4, Max Generation Length=80962026.0575.39——
EOPDDistillation Domain=Single-Domain, Teacher Model=Skywork-OR1-Math-7B, Student Model=DeepSeek-R1-Distill-Qwen-1.5B, Training Steps=200, Learning Rate=5 x 10^-6, Sample Rollouts=4, Max Generation Length=80962026.0575.23——
GRPOBackbone=Qwen-3-1.7B-Base, Method=GRPO2026.037534.79—
Entropy OPD 20%Distillation Domain=Single-Domain, Teacher Model=Skywork-OR1-Math-7B, Student Model=DeepSeek-R1-Distill-Qwen-1.5B, Training Steps=200, Learning Rate=5 x 10^-6, Sample Rollouts=4, Max Generation Length=80962026.0573.82——
REOPOLD 2StageDistillation Domain=Single-Domain, Teacher Model=Skywork-OR1-Math-7B, Student Model=DeepSeek-R1-Distill-Qwen-1.5B, Training Steps=200, Learning Rate=5 x 10^-6, Sample Rollouts=4, Max Generation Length=80962026.0573.35——
DeepSeek-Qwen2.5-1.5BStudent Model=DeepSeek-Qwen2.5-1.5B2026.0571.01——
TrOPDDistillation Domain=Multi-Domain, Teacher Model=Skywork-OR1-7B, Student Model=DeepSeek-R1-Distill-Qwen-1.5B, Training Steps=200, Learning Rate=5 x 10^-6, Sample Rollouts=4, Max Generation Length=80962026.0570.93——
few-shotModel=OPEN-REASONER-7B2026.0565——
REOPOLDDistillation Domain=Multi-Domain, Teacher Model=Skywork-OR1-7B, Student Model=DeepSeek-R1-Distill-Qwen-1.5B, Training Steps=200, Learning Rate=5 x 10^-6, Sample Rollouts=4, Max Generation Length=80962026.0563.9——
OPDDistillation Domain=Multi-Domain, Teacher Model=Skywork-OR1-7B, Student Model=DeepSeek-R1-Distill-Qwen-1.5B, Training Steps=200, Learning Rate=5 x 10^-6, Sample Rollouts=4, Max Generation Length=80962026.0561.56——
LRSModel=OPEN-REASONER-7B, Prompting=0-shot2026.0560——
CoTModel=OPEN-REASONER-7B2026.0555——
BaseModel=OPEN-REASONER-7B, Prompting=0-shot2026.0550——
GRPO + ARROLBackbone=Llama-3.2-1B-Instruct, Method=GRPO + ARROL2026.0347.517.491.67
GRPOBackbone=Llama-3.2-1B-Instruct, Method=GRPO2026.034514.63—
LRS BASICModel=OPEN-REASONER-7B, Prompting=0-shot2026.0545——
LRSModel=OPEN-REASONER-1.5B, Prompting=0-shot2026.0537.5——
CoTModel=OPEN-REASONER-1.5B2026.0532.5——
LRS BASICModel=OPEN-REASONER-1.5B, Prompting=0-shot2026.0532.5——
Coupled-GRPOGeneration steps=30.0, Unmasking strategy=Static, Base Model Family=Dream2026.0530——
RLDFGeneration steps=256, Unmasking strategy=Static, Base Model Family=Dream2026.0530——
RLDFGeneration steps=512, Unmasking strategy=Static, Base Model Family=Dream2026.0530——
BaseModel=OPEN-REASONER-1.5B, Prompting=0-shot2026.0530——
few-shotModel=OPEN-REASONER-1.5B2026.0530——
Coupled-GRPOGeneration steps=512, Unmasking strategy=Static, Base Model Family=LLaDA2026.0522.5——
BaseBackbone=Llama3.1-8B-Instruct2026.0622.5——
RPOBackbone=Llama3.1-8B-Instruct2026.0622.5——
IPOBackbone=Llama3.1-8B-Instruct2026.0622.5——
TDPOBackbone=Llama3.1-8B-Instruct2026.0622.5——
RePO_detBackbone=Llama3.1-8B-Instruct2026.0622.5——
ESPOGeneration steps=512, Unmasking strategy=Static, Base Model Family=LLaDA2026.0520——
TraceRLGeneration steps=512, Unmasking strategy=Static, Base Model Family=LLaDA2026.0520——
d1Generation steps=128, Unmasking strategy=Static, Base Model Family=Dream2026.0520——
KTOBackbone=Llama3.1-8B-Instruct2026.0620——
d1Generation steps=256, Unmasking strategy=Static, Base Model Family=LLaDA2026.0517.5——
ESPOGeneration steps=256, Unmasking strategy=Static, Base Model Family=LLaDA2026.0517.5——
RLDFGeneration steps=128, Unmasking strategy=Static, Base Model Family=LLaDA2026.0517.5——
RLDFGeneration steps=512, Unmasking strategy=Static, Base Model Family=LLaDA2026.0517.5——
Dream-7B-InstructGeneration steps=256, Unmasking strategy=Static, Base Model Family=Dream2026.0517.5——
Dream-7B-InstructGeneration steps=512, Unmasking strategy=Static, Base Model Family=Dream2026.0517.5——
ESPOGeneration steps=512, Unmasking strategy=Static, Base Model Family=Dream2026.0517.5——
TraceRLGeneration steps=512, Unmasking strategy=Static, Base Model Family=Dream2026.0517.5——
Coupled-GRPOGeneration steps=256, Unmasking strategy=Static, Base Model Family=Dream2026.0517.5——
RLDFGeneration steps=128, Unmasking strategy=Static, Base Model Family=Dream2026.0517.5——
DPOBackbone=Llama3.1-8B-Instruct2026.0617.5——
LLaDA-8B-InstructGeneration steps=256, Unmasking strategy=Static, Base Model Family=LLaDA2026.0515——
LLaDA-1.5Generation steps=128, Unmasking strategy=Static, Base Model Family=LLaDA2026.0515——
d1Generation steps=512, Unmasking strategy=Static, Base Model Family=LLaDA2026.0515——
Coupled-GRPOGeneration steps=128, Unmasking strategy=Static, Base Model Family=LLaDA2026.0515——
RLDFGeneration steps=256, Unmasking strategy=Static, Base Model Family=LLaDA2026.0515——
Dream-7B-InstructGeneration steps=128, Unmasking strategy=Static, Base Model Family=Dream2026.0515——
d1Generation steps=256, Unmasking strategy=Static, Base Model Family=Dream2026.0515——
d1Generation steps=512, Unmasking strategy=Static, Base Model Family=Dream2026.0515——
ESPOGeneration steps=128, Unmasking strategy=Static, Base Model Family=Dream2026.0515——
ESPOGeneration steps=256, Unmasking strategy=Static, Base Model Family=Dream2026.0515——
TraceRLGeneration steps=256, Unmasking strategy=Static, Base Model Family=Dream2026.0515——
LLaDA-8B-InstructGeneration steps=512, Unmasking strategy=Static, Base Model Family=LLaDA2026.0512.5——
LLaDA-1.5Generation steps=256, Unmasking strategy=Static, Base Model Family=LLaDA2026.0512.5——
d1Generation steps=128, Unmasking strategy=Static, Base Model Family=LLaDA2026.0512.5——
Coupled-GRPOGeneration steps=128, Unmasking strategy=Static, Base Model Family=Dream2026.0512.5——
KTOBackbone=Llama3.1-8B-Base2026.0612.5——
LLaDA-1.5Generation steps=512, Unmasking strategy=Static, Base Model Family=LLaDA2026.0510——
ESPOGeneration steps=128, Unmasking strategy=Static, Base Model Family=LLaDA2026.0510——
Coupled-GRPOGeneration steps=256, Unmasking strategy=Static, Base Model Family=LLaDA2026.0510——
TraceRLGeneration steps=128, Unmasking strategy=Static, Base Model Family=Dream2026.0510——
DPOBackbone=Llama3.1-8B-Base2026.0610——
RePO_detBackbone=Llama3.1-8B-Base2026.0610——
LLaDA-8B-InstructGeneration steps=128, Unmasking strategy=Static, Base Model Family=LLaDA2026.057.5——
TraceRLGeneration steps=128, Unmasking strategy=Static, Base Model Family=LLaDA2026.057.5——
RPOBackbone=Llama3.1-8B-Base2026.067.5——
TraceRLGeneration steps=256, Unmasking strategy=Static, Base Model Family=LLaDA2026.055——
IPOBackbone=Llama3.1-8B-Base2026.065——
TDPOBackbone=Llama3.1-8B-Base2026.065——
BaseBackbone=Llama3.1-8B-Base2026.062.5——