List your company
ResearchFormerly WizWand

Mathematical Reasoning on MATH (Exact Match Accuracy)

88.4AccuracyPrefix-RFT*
-3.53620.33244.268.068Oct 30, 2025Nov 27, 2025Dec 25, 2025Jan 22, 2026Feb 19, 2026Mar 19, 2026Apr 16, 2026
Updated 6mo ago

Evaluation results

MethodLinks
Prefix-RFT*Backbone=Qwen2.5-Math-7B, Reported from original paper=true2025.1088.4
ICPOBackbone=Qwen2.5-Math-7B2025.1088.4
LUFFY*Backbone=Qwen2.5-Math-7B, Reported from original paper=true2025.1087.6
ICPO†Backbone=Qwen2.5-Math-7B2025.1086.6
SFT+RL_GRPOBackbone=Qwen2.5-Math-7B2025.1085.8
ReLIFT*Backbone=Qwen2.5-Math-7B, Reported from original paper=true2025.1085
OpenReasoner-Zero*Backbone=Qwen2.5-Math-7B, Reported from original paper=true2025.1082.4
SFTBackbone=Qwen2.5-Math-7B2025.1082.2
PRIME-Zero*Backbone=Qwen2.5-Math-7B, Reported from original paper=true2025.1081.4
RL_GRPOBackbone=Qwen2.5-Math-7B2025.1081.2
Oat-Zero*Backbone=Qwen2.5-Math-7B, Reported from original paper=true2025.1078
SimpleRL-Zero*Backbone=Qwen2.5-Math-7B, Reported from original paper=true2025.1076
Qwen2.5-7B-InstructBase Model=Qwen2.5-7B-Instruct2026.0476
FineSteerBase Model=Qwen2.5-7B-Instruct2026.0475
AlphaSteerBase Model=Qwen2.5-7B-Instruct2026.0474
FineSteerBase Model=Qwen2.5-7B-Instruct, Gating Training Samples=20002026.0474
AlphaSteerBase Model=Llama-3.1-8B-Instruct2026.0455
FineSteerBase Model=Llama-3.1-8B-Instruct, Gating Training Samples=20002026.0454
AlphaSteerBase Model=Llama-3.1-8B-Instruct, Gating Training Samples=20002026.0453
FineSteerBase Model=Llama-3.1-8B-Instruct2026.0452
Llama-3.1-8B-InstructBase Model=Llama-3.1-8B-Instruct2026.0451
AlphaSteerBase Model=Qwen2.5-7B-Instruct, Gating Training Samples=20002026.0448
Qwen2.5-Math-7BBackbone=Qwen2.5-Math-7B2025.1043.6
UniPROTSelection Protocol=source-wise2026.0438.4
GREATSSelection Protocol=source-wise2026.0437.84
UniPROTSelection Protocol=batch-wise2026.0437.76
COLMSelection Protocol=source-wise2026.0437.28
GREATSSelection Protocol=batch-wise2026.0437.28
FTBase Model=PHI-32026.0436.54
COLMSelection Protocol=batch-wise2026.0436.42
GradNormSelection Protocol=source-wise2026.0436.1
BiPOBase Model=Qwen2.5-7B-Instruct2026.0436
SBERTSelection Protocol=source-wise2026.0435.54
SBERTSelection Protocol=batch-wise2026.0435.06
GradNormSelection Protocol=batch-wise2026.0435.03
MaxLossSelection Protocol=source-wise2026.0432.05
MaxLossSelection Protocol=batch-wise2026.0430.02
TruthFlowBase Model=Llama-3.1-8B-Instruct2026.0428
BiPOBase Model=Llama-3.1-8B-Instruct2026.043
TruthFlowBase Model=Qwen2.5-7B-Instruct2026.040