List your company
ResearchFormerly WizWand

General Knowledge Reasoning on MMLU High

67.2AccuracyReason, Reward, Refine
29.115239.002648.8958.7774Jul 6, 2026
Updated 4mo ago

Evaluation results

MethodLinks
Reason, Reward, RefineModel=LLaMA 3.2 3B2026.0767.2
Reason, Reward, RefineModel=Phi 3.5 Mini 3.8B2026.0765.83
RAGModel=Phi 3.5 Mini 3.8B2026.0760.53
RAGModel=Qwen 2.5 1.5B2026.0759.41
DPOModel=Phi 3.5 Mini 3.8B2026.0756.12
SFTModel=Phi 3.5 Mini 3.8B2026.0755.97
Reason, Reward, RefineModel=LLaMA 3.2 1B2026.0755.29
RAGModel=LLaMA 3.2 3B2026.0755.29
CoTModel=Phi 3.5 Mini 3.8B2026.0755.17
Reason, Reward, RefineModel=Qwen 2.5 1.5B2026.0755.03
CoTModel=LLaMA 3.2 3B2026.0750
SFTModel=LLaMA 3.2 3B2026.0748.29
DPOModel=LLaMA 3.2 3B2026.0747.65
CoTModel=Qwen 2.5 1.5B2026.0747.05
DPOModel=Qwen 2.5 1.5B2026.0746.47
SFTModel=LLaMA 3.2 1B2026.0742.35
SFTModel=Qwen 2.5 1.5B2026.0741.18
DPOModel=LLaMA 3.2 1B2026.0739.41
RAGModel=LLaMA 3.2 1B2026.0738.02
CoTModel=LLaMA 3.2 1B2026.0730.58