List your company
ResearchFormerly WizWand

Mathematical Reasoning on AMC (Pass@1 accuracy)

83.75Pass@1 AccuracySFPO
0.404422.042243.6865.3178Sep 24, 2025Nov 4, 2025Dec 15, 2025Jan 25, 2026Mar 7, 2026Apr 17, 2026May 28, 2026
Updated 5mo ago

Evaluation results

MethodLinks
SFPOModel=DS-distilled-Qwen-7B2025.1083.75
GRPOModel=DS-distilled-Qwen-7B2025.1080.42
SFPOModel=Qwen2.5-Math-7B2025.1074.49
GRPOModel=Qwen2.5-Math-7B2025.1071.08
SFPOModel=DS-distilled-Qwen-1.5B2025.1070.28
SCRLCandidate responses=64, Training samples=32, Backbone=Qwen2.5-Math-7B2026.0368.5
TTRLCandidate responses=64, Training samples=32, Backbone=Qwen2.5-Math-7B2026.0368.1
SCRLCandidate responses=32, Training samples=16, Backbone=Qwen2.5-Math-7B2026.0366.9
GRPOModel=DS-distilled-Qwen-1.5B2025.1066.86
TTRLCandidate responses=32, Training samples=16, Backbone=Qwen2.5-Math-7B2026.0365.7
BaseModel=DS-distilled-Qwen-7B2025.1061.75
PrismBase model=Qwen3-4B-Base2026.0361.25
GRPOModel=Qwen3-4B-Base2025.1059.03
SPICEBase model=Qwen3-4B-Base2026.0357.5
R-ZeroBase model=Qwen3-4B-Base2026.0357.27
SFPOModel=Qwen3-4B-Base2025.1057.23
SRGenModel=Qwen3-32B2025.1056.8
SFPOModel=Qwen2.5-Math-1.5B2025.1056.02
Self-RefineModel=Qwen3-32B2025.1054.4
CoTModel=Qwen3-32B2025.1054
BaseModel=Qwen2.5-Math-7B2025.1053.61
GRPOModel=Qwen2.5-Math-1.5B2025.1053.31
R-FewBase model=Qwen3-4B-Base, Selection percentage=1%2026.0352.7
Absolute ZeroBase model=Qwen3-4B-Base2026.0352.5
R-FewBase model=Qwen3-4B-Base, Selection percentage=5%2026.0352.4
DCRLBackbone=Qwen3-8B-Base2026.0351.9
Self-RefineModel=DS-R1-Qwen-7B2025.1051.6
SRGenModel=DS-R1-Qwen-7B2025.1051.2
CoTModel=DS-R1-Qwen-7B2025.1051
SRGenModel=DS-R1-Llama-8B2025.1050.6
CoTModel=DS-R1-Llama-8B2025.1050
Self-RefineModel=DS-R1-Llama-8B2025.1049.4
BaseModel=DS-distilled-Qwen-1.5B2025.1047.59
Qwen3-4B-Base2026.0347.5
VeriGatePolicy Model=Qwen2.5-7B-Instruct, PRM Model=Qwen-PRM-7B2026.0545.78
DCRLBackbone=Qwen3-4B-Base2026.0345.6
KTOBackbone=Qwen2.5-Math-7B2025.0943.1
FPABackbone=Qwen3-4B-Base2025.0942.8
FPABackbone=Qwen2.5-Math-7B2025.0942.3
GRPO-VanillaPolicy Model=Qwen2.5-7B-Instruct, PRM Model=Qwen-PRM-7B2026.0542.17
Off-RLBackbone=Qwen2.5-Math-7B2025.0942
SCRLCandidate responses=32, Training samples=16, Backbone=Qwen2.5-3B2026.0341.5
SCRLCandidate responses=64, Training samples=32, Backbone=Qwen2.5-3B2026.0341.3
SRGenModel=Qwen2.5-Math-7B2025.1041.2
PRM-as-ORMPolicy Model=Qwen2.5-7B-Instruct, PRM Model=Qwen-PRM-7B2026.0540.96
SFTBackbone=Qwen2.5-Math-7B2025.0940.8
RPOBackbone=Qwen2.5-Math-7B2025.0940.7
DPOBackbone=Qwen2.5-Math-7B2025.0940
TTRLCandidate responses=64, Training samples=32, Backbone=Qwen2.5-3B2026.0339.5
TTRLCandidate responses=32, Training samples=16, Backbone=Qwen2.5-3B2026.0339.4
DPOBackbone=Qwen3-4B-Base2025.0939.4
A*-POBackbone=Qwen2.5-Math-7B2025.0938.9
BaseModel=Qwen2.5-Math-1.5B2025.1037.65
Off-RLBackbone=Qwen3-4B-Base2025.0936.9
BaseBackbone=Qwen2.5-Math-7B2025.0936.8
KTOBackbone=Qwen3-4B-Base2025.0935.4
Qwen2.5-Math-7BCandidate responses=32, Training samples=16, Backbone=Qwen2.5-Math-7B2026.0334
Qwen2.5-Math-7BCandidate responses=64, Training samples=32, Backbone=Qwen2.5-Math-7B2026.0334
CoTModel=Qwen2.5-Math-7B2025.1034
RPOBackbone=Qwen3-4B-Base2025.0933.5
A*-POBackbone=Qwen3-4B-Base2025.0933.3
Self-RefineModel=Qwen2.5-Math-7B2025.1032.4
SFTBackbone=Qwen3-4B-Base2025.0931.2
BaseBackbone=Qwen3-4B-Base2025.0929.8
VeriGatePolicy Model=Qwen2.5-1.5B-Instruct, PRM Model=Qwen-PRM-7B2026.0527.71
PRM-as-ORM (+Gate)Policy Model=Qwen2.5-1.5B-Instruct, PRM Model=Qwen-PRM-7B2026.0525.3
DCRLBackbone=Llama3.2-3B-Instruct2026.0323.3
Qwen2.5-3BCandidate responses=32, Training samples=16, Backbone=Qwen2.5-3B2026.0323.2
Qwen2.5-3BCandidate responses=64, Training samples=32, Backbone=Qwen2.5-3B2026.0323.2
VeriGate (Simplified)Policy Model=Qwen2.5-1.5B-Instruct, PRM Model=Qwen-PRM-7B2026.0521.69
BaseModel=Qwen3-4B-Base2025.1020.48
FPABackbone=Llama-3.2-3B-Instruct2025.0918.4
BaseBackbone=Llama-3.2-3B-Instruct2025.0916.3
SFTBackbone=Llama-3.2-3B-Instruct2025.0916.2
Dr.GRPOPolicy Model=Qwen2.5-1.5B-Instruct, PRM Model=Qwen-PRM-7B2026.0515.66
DPOBackbone=Llama-3.2-3B-Instruct2025.0915.6
GRPO-VanillaPolicy Model=Qwen2.5-1.5B-Instruct, PRM Model=Qwen-PRM-7B2026.0515.1
RPOBackbone=Llama-3.2-3B-Instruct2025.0914.9
KTOBackbone=Llama-3.2-3B-Instruct2025.0914.8
Off-RLBackbone=Llama-3.2-3B-Instruct2025.0914.1
DAPOPolicy Model=Qwen2.5-1.5B-Instruct, PRM Model=Qwen-PRM-7B2026.0514.1
A*-POBackbone=Llama-3.2-3B-Instruct2025.0913.4
VeriGate (-Gate)Policy Model=Qwen2.5-1.5B-Instruct, PRM Model=Qwen-PRM-7B2026.057.23
PRM-as-ORMPolicy Model=Qwen2.5-1.5B-Instruct, PRM Model=Qwen-PRM-7B2026.053.61