List your company
ResearchFormerly WizWand

Mathematical Reasoning on AIME 2024 (avg@10)

15.33Avg@10 RecallSFT + GRPO
1.46685.06598.66512.2641Jan 12, 2026
Updated 8mo ago

Evaluation results

MethodLinks
SFT + GRPOBackbone=Qwen2.5-7B-Instruct2026.0115.33
SCR (Ours)Backbone=Qwen2.5-7B-Instruct2026.0113.67
GRPOBackbone=Qwen2.5-7B-Instruct2026.0111.33
SCR-Stage IBackbone=Qwen2.5-7B-Instruct2026.0111.33
BaseBackbone=Qwen2.5-7B-Instruct2026.0111
Self-RefineBackbone=Qwen2.5-7B-Instruct2026.019.33
SCR-SFTBackbone=Qwen2.5-7B-Instruct2026.019.33
GRPOBackbone=Llama3.1-8B-Instruct2026.018.33
SFT + GRPOBackbone=Qwen2.5-3B-Instruct2026.017.33
SCR (Ours)Backbone=Qwen2.5-3B-Instruct2026.017.33
BaseBackbone=Qwen2.5-3B-Instruct2026.017
Self-RefineBackbone=Qwen2.5-3B-Instruct2026.016.67
GRPOBackbone=Qwen2.5-3B-Instruct2026.016
SFT + GRPOBackbone=Llama3.1-8B-Instruct2026.016
SCR (Ours)Backbone=Llama3.1-8B-Instruct2026.015.33
SCR-Stage IBackbone=Qwen2.5-3B-Instruct2026.014.33
SCR-Stage IBackbone=Llama3.1-8B-Instruct2026.014
SCR-SFTBackbone=Llama3.1-8B-Instruct2026.013
SCR-SFTBackbone=Qwen2.5-3B-Instruct2026.012.33
BaseBackbone=Llama3.1-8B-Instruct2026.012
Self-RefineBackbone=Llama3.1-8B-Instruct2026.012