List your company
ResearchFormerly WizWand

Mathematical Reasoning on MATH OOD

80.2AccuracyGPT-4.5
-0.50420.44841.462.352Jan 28, 2026Feb 18, 2026Mar 11, 2026Apr 1, 2026Apr 22, 2026May 13, 2026Jun 3, 2026
Updated 5mo ago

Evaluation results

MethodLinks
GPT-4.52026.0680.2
IGA (Ours)2026.0671.6
Pareto-GS2026.0665.8
SAND-Mask2026.0662.9
PCGrad2026.0661.8
V-REx2026.0661.3
IRM2026.0660.8
ERM-SFT2026.0658.4
CoT-Distill2026.0657.9
LoRA-SFT2026.0657.3
OGERBackbone=Qwen2.5-Math-7B2026.0451.61
LuffyBackbone=Qwen2.5-Math-7B2026.0451.33
OGERBackbone=Qwen2.5-Math-1.5B2026.0437.7
SFT+GRPOBackbone=Qwen-2.5-0.5B, Fine-tuning strategy=Full2026.0134.53
LuffyBackbone=Qwen2.5-Math-1.5B2026.0433.07
SAPO-iter3Backbone=Qwen-2.5-0.5B, Fine-tuning strategy=Full2026.0131.72
SAPO-iter2Backbone=Qwen-2.5-0.5B, Fine-tuning strategy=Full2026.0131.48
GRPOBackbone=Qwen2.5-Math-7B2026.0431.21
SAPO-iter1Backbone=Qwen-2.5-0.5B, Fine-tuning strategy=Full2026.0130.04
Online-RFTBackbone=Qwen-2.5-0.5B, Fine-tuning strategy=Full2026.0128.85
RFTBackbone=Qwen-2.5-0.5B, Fine-tuning strategy=Full2026.0127.35
BaseBackbone=Qwen2.5-Math-7B2026.0426.28
CoTBackbone=Qwen-2.5-0.5B, Fine-tuning strategy=Full2026.0125.97
SFTBackbone=Qwen-2.5-0.5B, Fine-tuning strategy=Full2026.0124.84
Online-RFTBackbone=Gemma-2-2B, Fine-tuning strategy=QLoRA2026.0124.6
SFT+GRPOBackbone=Gemma-2-2B, Fine-tuning strategy=QLoRA2026.0124.36
SAPO-iter2Backbone=Gemma-2-2B, Fine-tuning strategy=QLoRA2026.0124.36
SAPO-iter3Backbone=Gemma-2-2B, Fine-tuning strategy=QLoRA2026.0124.24
SAPO-iter1Backbone=Gemma-2-2B, Fine-tuning strategy=QLoRA2026.0124
RPOBackbone=Qwen-2.5-0.5B, Fine-tuning strategy=Full2026.0122.68
RFTBackbone=Gemma-2-2B, Fine-tuning strategy=QLoRA2026.0122.14
BaseBackbone=Qwen2.5-Math-1.5B2026.0421.32
SFTBackbone=Gemma-2-2B, Fine-tuning strategy=QLoRA2026.0120.95
CoTBackbone=Gemma-2-2B, Fine-tuning strategy=QLoRA2026.0116.22
RFT+DPOBackbone=Gemma-2-2B, Fine-tuning strategy=QLoRA2026.0115.38
GRPOBackbone=Qwen2.5-Math-1.5B2026.0415.37
RFT+DPOBackbone=Qwen-2.5-0.5B, Fine-tuning strategy=Full2026.0112.45
RPOBackbone=Gemma-2-2B, Fine-tuning strategy=QLoRA2026.0111.96
SFT+GRPOBackbone=Llama-3.2-1B, Fine-tuning strategy=Full2026.015.92
SAPO-iter2Backbone=Llama-3.2-1B, Fine-tuning strategy=Full2026.015.92
RFTBackbone=Llama-3.2-1B, Fine-tuning strategy=Full2026.015.56
SAPO-iter1Backbone=Llama-3.2-1B, Fine-tuning strategy=Full2026.015.51
Online-RFTBackbone=Llama-3.2-1B, Fine-tuning strategy=Full2026.015.45
SAPO-iter3Backbone=Llama-3.2-1B, Fine-tuning strategy=Full2026.015.45
RPOBackbone=Llama-3.2-1B, Fine-tuning strategy=Full2026.014.85
RFT+DPOBackbone=Llama-3.2-1B, Fine-tuning strategy=Full2026.014.06
CoTBackbone=Llama-3.2-1B, Fine-tuning strategy=Full2026.013.65
SFTBackbone=Llama-3.2-1B, Fine-tuning strategy=Full2026.012.6