List your company
ResearchFormerly WizWand

Code Reasoning on CRUX

87.37AccuracyRMoA
18.865236.650154.43572.2199May 26, 2025Jul 23, 2025Sep 19, 2025Nov 17, 2025Jan 14, 2026Mar 13, 2026May 11, 2026
Updated 5mo ago

Evaluation results

MethodLinks
RMoAModel=GPT-4o2025.0587.37
SMoAModel=GPT-4o2025.0586.93
MoAModel=GPT-4o2025.0586.66
GPT-4oModel=GPT-4o2025.0575.8
RMoAModel=Qwen2.5-7B-Instruct2025.0561
SMoAModel=Qwen2.5-7B-Instruct2025.0559.93
Qwen2.5-7B-InstructModel=Qwen2.5-7B-Instruct2025.0557.31
MoAModel=Qwen2.5-7B-Instruct2025.0556.81
MoAModel=Gemma2-9B-Instruct2025.0551.5
SMoAModel=Gemma2-9B-Instruct2025.0551.25
RMoAModel=Gemma2-9B-Instruct2025.0550.5
Gemma2-9B-InstructModel=Gemma2-9B-Instruct2025.0547.5
MoAModel=Llama3.1-8B-Instruct2025.0546.12
SMoAModel=Llama3.1-8B-Instruct2025.0544.81
RMoAModel=Llama3.1-8B-Instruct2025.0542.65
Llama3.1-8B-InstructModel=Llama3.1-8B-Instruct2025.0540.62
INTUITORBackbone=Qwen2.5-3B-Base, Training Dataset=CodeContests, RL Strategy=INTUITOR2026.0539.38
VIGORBackbone=Qwen2.5-3B-Base, Training Dataset=CodeContests, RL Strategy=VIGOR2026.0535.62
INTUITORModel=Llama3.2-3B-Ins2025.0529.3
GRPO-PVModel=Llama3.2-3B-Ins2025.0528.1
GRPOModel=Llama3.2-3B-Ins2025.0526.6
BaselineModel=Llama3.2-3B-Ins2025.0526.5
Before RL (Base)Backbone=Qwen2.5-3B-Base, Training Dataset=CodeContests, RL Strategy=None2026.0524.38
BaselineModel=OLMo2-7B-SFT2025.0523.8
GRPOModel=OLMo2-7B-SFT2025.0521.8
INTUITORModel=OLMo2-7B-SFT2025.0521.5