List your company
ResearchFormerly WizWand

Mathematical Reasoning on AMC (pass@32, Overall)

85.3Average Pass@32Qwen3
27.68442.64257.672.558Apr 13, 2026Apr 18, 2026Apr 23, 2026Apr 28, 2026May 3, 2026May 8, 2026May 13, 2026
Updated 5mo ago

Evaluation results

MethodLinks
Qwen3Model Category=Original Models2026.0585.3——
ConSPOBackbone=DeepSeek-R1-Distill-Qwen-7B2026.0583.8——
GRPOBackbone=DeepSeek-R1-Distill-Qwen-7B2026.0582.8——
Dr. GRPOBackbone=DeepSeek-R1-Distill-Qwen-7B2026.0582.2——
DisCOBackbone=DeepSeek-R1-Distill-Qwen-7B2026.0582——
ConSPOBackbone=DeepSeek-R1-Distill-Llama-8B2026.0581.1——
DAPOBackbone=DeepSeek-R1-Distill-Qwen-7B2026.0580.6——
Dr. GRPOBackbone=DeepSeek-R1-Distill-Llama-8B2026.0580.5——
GRPOBackbone=DeepSeek-R1-Distill-Llama-8B2026.0578.6——
DisCOBackbone=DeepSeek-R1-Distill-Llama-8B2026.0578.4——
DAPOBackbone=DeepSeek-R1-Distill-Llama-8B2026.0577.6——
EAPOMethod Category=Policy Entropy Method, Backbone=DeepScaleR-1.5B-8k2026.0475.595.557
KL-CovMethod Category=Policy Entropy Method, Backbone=DeepScaleR-1.5B-8k2026.04759156.5
EntroAdvMethod Category=Policy Entropy Method, Backbone=DeepScaleR-1.5B-8k2026.0473.992.956.3
DAPOMethod Category=Classic RLVR Method, Backbone=DeepScaleR-1.5B-8k2026.0473.690.855.1
GRPOMethod Category=Classic RLVR Method, Backbone=DeepScaleR-1.5B-8k2026.0471.492.453.9
ForkingMethod Category=Policy Entropy Method, Backbone=DeepScaleR-1.5B-8k2026.047190.753.7
ConSPOBase Model=DeepSeek-R1-Distill-Qwen-1.5B2026.0570.4——
Base ModelBackbone=DeepSeek-R1-Distill-Qwen-7B2026.0570.3——
EntroRegMethod Category=Policy Entropy Method, Backbone=DeepScaleR-1.5B-8k2026.0470.290.651.9
SAPOBase Model=DeepSeek-R1-Distill-Qwen-1.5B2026.0570.1——
GMPOBase Model=DeepSeek-R1-Distill-Qwen-1.5B2026.0570——
W-REINFMethod Category=Classic RLVR Method, Backbone=DeepScaleR-1.5B-8k2026.0469.89151.8
Base-1.5BMethod Category=Classic RLVR Method, Backbone=DeepScaleR-1.5B-8k2026.0469.489.950.7
DisCOBase Model=DeepSeek-R1-Distill-Qwen-1.5B2026.0566.5——
DAPOBase Model=DeepSeek-R1-Distill-Qwen-1.5B2026.0566.2——
CISPOBase Model=DeepSeek-R1-Distill-Qwen-1.5B2026.0565.2——
GRPOBase Model=DeepSeek-R1-Distill-Qwen-1.5B2026.0564.4——
Base ModelBackbone=DeepSeek-R1-Distill-Llama-8B2026.0563.5——
Dr. GRPOBase Model=DeepSeek-R1-Distill-Qwen-1.5B2026.0563.5——
Oat-ZeroModel Category=On-Policy Methods2026.0561.2——
TGPO-annealingModel Category=On-Policy Methods2026.0560.2——
TGPOModel Category=On-Policy Methods2026.0558.6——
TGPORModel Category=On-Policy Methods2026.0558.4——
GRPO++Model Category=On-Policy Methods2026.0558.3——
LUFFYModel Category=Off-Policy and Mixed-Policy2026.0557.6——
KDRLModel Category=On-Policy Methods2026.0555.8——
SimpleRL-ZeroModel Category=On-Policy Methods2026.0554.9——
PRIME-ZeroModel Category=On-Policy Methods2026.0554——
Base ModelBase Model=DeepSeek-R1-Distill-Qwen-1.5B2026.0552.8——
Qwen3-8192Model Category=Original Models, Max Generation Length=81922026.0552.2——
SFTModel Category=Off-Policy and Mixed-Policy2026.0545.3——
Qwen2.5-Math-7BModel Category=Original Models2026.0531.3——
OP DistillModel Category=On-Policy Methods2026.0529.9——
AV-GRPOAlgorithm=AV-GRPO, Backbone=Qwen2.5-Math-1.5B2026.03—81.950.9
BaseBackbone=Qwen2.5-7B-Instruct2025.08—52.58—
BaseBackbone=Llama3.1-8B-Instruct2025.08—24.45—
CDA-GRPOAlgorithm=CDA-GRPO, Backbone=Qwen2.5-Math-1.5B2026.03—85.549
FA-GRPOAlgorithm=FA-GRPO, Backbone=Qwen2.5-Math-1.5B2026.03—83.149.9
GRPOAlgorithm=GRPO, Backbone=Qwen2.5-Math-1.5B2026.03—84.347.8
Llama3.1-8B-InstructTraining Pipeline=SFT2025.08—47.19—
Llama3.1-8B-InstructTraining Pipeline=SFT → GRPO2025.08—54.38—
Llama3.1-8B-InstructTraining Pipeline=PSFT2025.08—46.8—
Llama3.1-8B-InstructTraining Pipeline=PSFT → GRPO2025.08—55.23—
MSA-GRPOAlgorithm=MSA-GRPO, Backbone=Qwen2.5-Math-1.5B2026.03—84.350.7
PR-GRPOAlgorithm=PR-GRPO, Backbone=Qwen2.5-Math-1.5B2026.03—80.749.4
PSFTBackbone=Qwen2.5-7B-Instruct2025.08—62.34—
PSFTBackbone=Llama3.1-8B-Instruct2025.08—46.8—
PSFTwarm-upBackbone=Qwen2.5-7B-Instruct2025.08—62.42—
PSFTwarm-upBackbone=Llama3.1-8B-Instruct2025.08—49.45—
Qwen2.5-7B-InstructTraining Pipeline=SFT2025.08—62.73—
Qwen2.5-7B-InstructTraining Pipeline=SFT → GRPO2025.08—70.86—
Qwen2.5-7B-InstructTraining Pipeline=PSFT2025.08—62.34—
Qwen2.5-7B-InstructTraining Pipeline=PSFT → GRPO2025.08—71.72—
SFTBackbone=Qwen2.5-7B-Instruct2025.08—62.73—
SFTBackbone=Llama3.1-8B-Instruct2025.08—47.19—
SFT-KLBackbone=Qwen2.5-7B-Instruct2025.08—63.2—
SFT-KLBackbone=Llama3.1-8B-Instruct2025.08—45.55—
SVE-LNA-GRPOAlgorithm=SVE-LNA-GRPO, Backbone=Qwen2.5-Math-1.5B2026.03—89.248.7
VM-AV-GRPOAlgorithm=VM-AV-GRPO, Backbone=Qwen2.5-Math-1.5B2026.03—8852.5