List your company
ResearchFormerly WizWand

Scientific Reasoning on JEEBench

57.34Mean AccuracyReason, Reward, Refine
27.304835.102442.950.6976Apr 13, 2026Apr 27, 2026May 11, 2026May 25, 2026Jun 8, 2026Jun 22, 2026Jul 6, 2026
Updated 3mo ago

Evaluation results

MethodLinks
Reason, Reward, RefineBackbone=LLaMA 3.2 3B2026.0757.34
Reason, Reward, RefineBackbone=Qwen 2.5 1.5B2026.0754.86
Qwen2.5 32B + RL (synthetic)Model Size=32B, Post-training=RL (synthetic)2026.0452.28
Reason, Reward, RefineBackbone=Phi 3.5 Mini 3.8B2026.0750.23
Reason, Reward, RefineBackbone=LLaMA 3.2 1B2026.0749.17
DPOBackbone=Phi 3.5 Mini 3.8B2026.0741.65
RAGBackbone=Phi 3.5 Mini 3.8B2026.0741.25
RAGBackbone=LLaMA 3.2 3B2026.0740.83
RAGBackbone=Qwen 2.5 1.5B2026.0740
FTBackbone=Qwen 2.5 1.5B2026.0739.02
DPOBackbone=LLaMA 3.2 3B2026.0738.21
FTBackbone=Phi 3.5 Mini 3.8B2026.0738.04
CoTBackbone=Qwen 2.5 1.5B2026.0737.53
FTBackbone=LLaMA 3.2 3B2026.0736.59
CoTBackbone=Phi 3.5 Mini 3.8B2026.0735.9
DPOBackbone=LLaMA 3.2 1B2026.0735.77
RAGBackbone=LLaMA 3.2 1B2026.0735.5
Qwen2.5 32BModel Size=32B, Variant=Instruct2026.0434.38
CoTBackbone=LLaMA 3.2 1B2026.0732.52
CoTBackbone=LLaMA 3.2 3B2026.0730.21
DPOBackbone=Qwen 2.5 1.5B2026.0730.08
FTBackbone=LLaMA 3.2 1B2026.0728.46