List your company
ResearchFormerly WizWand

Mathematical Reasoning on MATH-500 (Testing accuracy (%))

92.6Accuracy (MATH-500)MILES
41.22454.56267.981.238Sep 28, 2025Nov 14, 2025Dec 31, 2025Feb 16, 2026Apr 4, 2026May 21, 2026Jul 8, 2026
Updated 3mo ago

Evaluation results

MethodLinks
MILESBackbone=GPT-4.1-mini, Inference Strategy=MILES2026.0792.6
MILESBackbone=GPT-4.1, Inference Strategy=MILES2026.0791.6
SCBackbone=GPT-4.1-mini, Inference Strategy=SC2026.0789.4
ZS-CoTBackbone=GPT-4.1-mini, Inference Strategy=ZS-CoT2026.0788
SCBackbone=GPT-4.1, Inference Strategy=SC2026.0788
ZS-CoTBackbone=GPT-4.1, Inference Strategy=ZS-CoT2026.0787.2
DCBackbone=GPT-4.1-mini, Inference Strategy=DC2026.0787
BoTBackbone=GPT-4.1-mini, Inference Strategy=BoT2026.0786.8
BoTBackbone=GPT-4.1, Inference Strategy=BoT2026.0784.8
DCBackbone=GPT-4.1, Inference Strategy=DC2026.0783.2
GAPGModel=Qwen2.5-1.5B2025.0981.6
GeoMinBackbone=Deepseek-R1-Distill-Qwen-1.5B, Supervision Setting=semi-supervised (10% labeled data), Domain Context=In-Domain2026.0679.6
TraPOBackbone=Deepseek-R1-Distill-Qwen-1.5B, Supervision Setting=semi-supervised (10% labeled data), Domain Context=In-Domain2026.0678.2
TTRLBackbone=Deepseek-R1-Distill-Qwen-1.5B, Supervision Setting=semi-supervised (10% labeled data), Domain Context=In-Domain2026.0677.6
Co-rewardingBackbone=Deepseek-R1-Distill-Qwen-1.5B, Supervision Setting=semi-supervised (10% labeled data), Domain Context=In-Domain2026.0677.6
Self-certaintyBackbone=Deepseek-R1-Distill-Qwen-1.5B, Supervision Setting=semi-supervised (10% labeled data), Domain Context=In-Domain2026.0675.6
HIPPOTraining Data=DeepScaleR2026.0674.6
Router-R1Model=Qwen2.5-1.5B2025.0974.1
Seq-entropyBackbone=Deepseek-R1-Distill-Qwen-1.5B, Supervision Setting=semi-supervised (10% labeled data), Domain Context=In-Domain2026.0672.4
Full dataTraining regime=Full dataset2026.0572.2
RFGOOptimization=Reasoning Feature-Guided Optimization2026.0572
Tok-entropyBackbone=Deepseek-R1-Distill-Qwen-1.5B, Supervision Setting=semi-supervised (10% labeled data), Domain Context=In-Domain2026.0670.2
GAPGModel=LLaMA-3.2-3B2025.0968.6
AugHard@8Training regime=Rewritten AugHard@8 subset2026.0568.4
hard@8Training regime=hard@8 subset2026.0565
Router-R1Model=LLaMA-3.2-3B2025.0964.9
Qwen3-4B + SFTTraining Data=DeepScaleR2026.0660.2
AutoMixModel=Qwen2.5-1.5B2025.0957.2
ZO Fine-tunerBackbone=Qwen2.5-14B, Fine-tuning dataset=MetaMathQA2025.1054.6
MeZOBackbone=Qwen2.5-14B, Fine-tuning dataset=MetaMathQA2025.1053
Qwen3-4BTraining Data=DeepScaleR2026.0649.6
AutoMixModel=LLaMA-3.2-3B2025.0947.9
Base ModelBackbone=Qwen2.5-14B2025.1043.2