List your company
ResearchFormerly WizWand

Mathematical Reasoning on DeepMind-Mathematics

88.4AccuracyQwen3-I(4B)
10.50430.72750.9571.173Jun 18, 2024Oct 15, 2024Feb 12, 2025Jun 12, 2025Oct 9, 2025Feb 6, 2026Jun 6, 2026
Updated 4mo ago

Evaluation results

MethodLinks
Qwen3-I(4B)Model Family=Qwen3, Target Model (TL)=ℐ (4B), Unlock/Instruction-tuning Configuration (SU)=-2026.0488.4
Ministral-3-I(14B)Model Family=Ministral-3, Target Model (TL)=ℐ (14B), Unlock/Instruction-tuning Configuration (SU)=-2026.0487.2
Ministral-3-I(3B)Model Family=Ministral-3, Target Model (TL)=ℐ (3B), Unlock/Instruction-tuning Configuration (SU)=-2026.0484.2
UNLOCKModel Family=Qwen3, Target Model (TL)=14B, Unlock/Instruction-tuning Configuration (SU)=+UNLOCK from 4B, Transfer Setting=Second Setting (Red Shading)2026.0482.4
Qwen3-I(14B)Model Family=Qwen3, Target Model (TL)=ℐ (14B), Unlock/Instruction-tuning Configuration (SU)=-2026.0480.1
UNLOCKModel Family=Qwen3, Target Model (TL)=14B, Unlock/Instruction-tuning Configuration (SU)=+UNLOCK from 4B, Transfer Setting=Task-Conditioned Transfer With Limited Data2026.0479.9
Qwen3-14BModel Family=Qwen3, Target Model (TL)=14B, Unlock/Instruction-tuning Configuration (SU)=-2026.0478.8
UNLOCKModel Family=Qwen3, Target Model (TL)=4B, Unlock/Instruction-tuning Configuration (SU)=+UNLOCK from 14B, Transfer Setting=Second Setting (Red Shading)2026.0476.5
UNLOCKModel Family=Qwen3, Target Model (TL)=4B, Unlock/Instruction-tuning Configuration (SU)=+UNLOCK from 14B, Transfer Setting=Task-Conditioned Transfer With Limited Data2026.0475.8
Qwen3-4BModel Family=Qwen3, Target Model (TL)=4B, Unlock/Instruction-tuning Configuration (SU)=-2026.0471.3
UNLOCKModel Family=Ministral-3, Target Model (TL)=8B, Unlock/Instruction-tuning Configuration (SU)=+UNLOCK from 3B, Transfer Setting=Task-Conditioned Transfer With Limited Data2026.0471.3
UNLOCKModel Family=Ministral-3, Target Model (TL)=8B, Unlock/Instruction-tuning Configuration (SU)=+UNLOCK from 3B, Transfer Setting=Second Setting (Red Shading)2026.0470.7
MathFusion-DSMath-7BBase Model=DeepSeekMath-7B, # Samples=195K2025.0369.1
Ministral-3-8BModel Family=Ministral-3, Target Model (TL)=8B, Unlock/Instruction-tuning Configuration (SU)=-2026.0467.4
UNLOCKModel Family=Ministral-3, Target Model (TL)=3B, Unlock/Instruction-tuning Configuration (SU)=+UNLOCK from 8B, Transfer Setting=Task-Conditioned Transfer With Limited Data2026.0466.2
MathFusion-DSMath-7BBase Model=DeepSeekMath-7B, # Samples=60K2025.0365.8
UNLOCKModel Family=Ministral-3, Target Model (TL)=3B, Unlock/Instruction-tuning Configuration (SU)=+UNLOCK from 8B, Transfer Setting=Second Setting (Red Shading)2026.0465.5
Ministral-3-3BModel Family=Ministral-3, Target Model (TL)=3B, Unlock/Instruction-tuning Configuration (SU)=-2026.0465.3
MathFusion-DSMath-7BBase Model=DeepSeekMath-7B, # Samples=30K, Training Strategy=Sequential2025.0364.6
DeepSeekMath-7B-DART-Math†Base Model=DeepSeekMath-7B, # Samples=60K2025.0362.8
MathFusion-DSMath-7BBase Model=DeepSeekMath-7B, # Samples=30K, Training Strategy=Parallel2025.0362.2
DeepSeekMath-7B-DART-MathBase Model=DeepSeekMath-7B, # Samples=590K2025.0361.6
DART-Math-DSMath-7BTraining Method=SFT, Data Selection Strategy=Prop2Diff, Backbone=DeepSeekMath-7B2024.0661.6
DeepSeekMath-7B-RFTBase Model=DeepSeekMath-7B, # Samples=590K2025.0360.2
DART-Math-DSMath-7BTraining Method=SFT, Data Selection Strategy=Uniform, Backbone=DeepSeekMath-7B2024.0660.2
DeepSeekMath-7B-RLTraining Method=RL, Backbone=DeepSeekMath-7B2024.0658.3
MathFusion-DSMath-7BBase Model=DeepSeekMath-7B, # Samples=30K, Training Strategy=Conditional2025.0355.2
DeepSeekMath-7B-MMIQCBase Model=DeepSeekMath-7B, # Samples=2.3M2025.0352.9
DeepSeekMath-7B-InstructBase Model=DeepSeekMath-7B, # Samples=780K2025.0352.2
Llama3-8B-DART-MathBase Model=Llama3-8B, # Samples=590K2025.0348
DeepSeekMath-7B-MetaMathBase Model=DeepSeekMath-7B, # Samples=60K2025.0345.9
Mistral-7B-DART-MathBase Model=Mistral-7B, # Samples=590K2025.0345.1
MathFusion-Llama3-8BBase Model=Llama3-8B, # Samples=60K2025.0343.4
MathFusion-Llama3-8BBase Model=Llama3-8B, # Samples=30K, Training Strategy=Sequential2025.0342
MathFusion-Llama3-8BBase Model=Llama3-8B, # Samples=30K, Training Strategy=Parallel2025.0341.9
Llama3-8B-RFTBase Model=Llama3-8B, # Samples=590K2025.0341.7
DeepSeekMath-7B-MMIQC†Base Model=DeepSeekMath-7B, # Samples=60K2025.0341.5
Llama3-8B-MMIQCBase Model=Llama3-8B, # Samples=2.3M2025.0341
Llama3-8B-DART-Math†Base Model=Llama3-8B, # Samples=60K2025.0339.9
MathFusion-Mistral-7BBase Model=Mistral-7B, # Samples=60K2025.0339.2
DeepSeekMath-7B-RefAugBase Model=DeepSeekMath-7B, # Samples=30K2025.0338.4
Mistral-7B-WizardMath-V1.1Base Model=Mistral-7B, # Samples=418K2025.0338.4
Mistral-7B-MMIQCBase Model=Mistral-7B, # Samples=2.3M2025.0338
GREATSBackbone=Qwen2.5-3B, Fine-tuning Dataset=Meta-MathQA, Subset Fraction=50%2026.0637
IDBackbone=Qwen2.5-3B, Fine-tuning Dataset=Meta-MathQA, Subset Fraction=50%2026.0637
IWDBackbone=Qwen2.5-3B, Fine-tuning Dataset=Meta-MathQA, Subset Fraction=50%2026.0636.75
Mistral-7B-DART-Math†Base Model=Mistral-7B, # Samples=60K2025.0336
COLMBackbone=Qwen2.5-3B, Fine-tuning Dataset=Meta-MathQA, Subset Fraction=50%2026.0635.7
Mistral-7B-RFTBase Model=Mistral-7B, # Samples=590K2025.0335.6
DeepSeekMath-7B-RefAugBase Model=DeepSeekMath-7B, # Samples=60K2025.0335.4
Llama3-8B-MetaMathBase Model=Llama3-8B, # Samples=400K2025.0335
PartitionSelBackbone=Qwen2.5-3B, Fine-tuning Dataset=Meta-MathQA, Subset Fraction=50%2026.0635
GradNormBackbone=Qwen2.5-3B, Fine-tuning Dataset=Meta-MathQA, Subset Fraction=50%2026.0633.4
RandomBackbone=Qwen2.5-3B, Fine-tuning Dataset=Meta-MathQA, Subset Fraction=50%2026.0632.81
Llama3-8B-MetaMath†Base Model=Llama3-8B, # Samples=60K2025.0331.3
Llama3-8B-MMIQC†Base Model=Llama3-8B, # Samples=60K2025.0330.9
MathFusion-Mistral-7BBase Model=Mistral-7B, # Samples=30K, Training Strategy=Sequential2025.0329.3
Llama3-8B-RefAugBase Model=Llama3-8B, # Samples=60K2025.0329.1
DeepSeekMath-7B-StandardBase Model=DeepSeekMath-7B, # Samples=15K2025.0328.6
Mistral-7B-MetaMathBase Model=Mistral-7B, # Samples=400K2025.0328
MathFusion-Llama3-8BBase Model=Llama3-8B, # Samples=30K, Training Strategy=Conditional2025.0327.4
Mistral-7B-MetaMath†Base Model=Mistral-7B, # Samples=60K2025.0327.2
MathFusion-Mistral-7BBase Model=Mistral-7B, # Samples=30K, Training Strategy=Parallel2025.0326.5
Llama3-8B-RefAugBase Model=Llama3-8B, # Samples=30K2025.0325.9
Llama3-8B-StandardBase Model=Llama3-8B, # Samples=15K2025.0321.6
MathFusion-Mistral-7BBase Model=Mistral-7B, # Samples=30K, Training Strategy=Conditional2025.0321.4
Mistral-7B-RefAugBase Model=Mistral-7B, # Samples=60K2025.0318.1
Mistral-7B-StandardBase Model=Mistral-7B, # Samples=15K2025.0317
COLMBase Model=Llama-3.2-3B, Training Dataset=MetaMathQA, Subset Ratio=12.5%2026.0616.6
COLMBackbone=Llama-3.1-8B, Training Dataset=MetaMathQA, Subset Ratio=12.5%2026.0616.6
PartitionSelBase Model=Llama-3.2-3B, Training Dataset=MetaMathQA, Subset Ratio=12.5%2026.0615.9
PartitionSelBackbone=Llama-3.1-8B, Training Dataset=MetaMathQA, Subset Ratio=12.5%2026.0615.9
GradNormBase Model=Llama-3.2-3B, Training Dataset=MetaMathQA, Subset Ratio=12.5%2026.0615.8
IDBase Model=Llama-3.2-3B, Training Dataset=MetaMathQA, Subset Ratio=12.5%2026.0615.8
GradNormBackbone=Llama-3.1-8B, Training Dataset=MetaMathQA, Subset Ratio=12.5%2026.0615.8
IDBackbone=Llama-3.1-8B, Training Dataset=MetaMathQA, Subset Ratio=12.5%2026.0615.8
Mistral-7B-RefAugBase Model=Mistral-7B, # Samples=30K2025.0315.4
Mistral-7B-MMIQC†Base Model=Mistral-7B, # Samples=60K2025.0313.5