List your company
ResearchFormerly WizWand

Mathematical Reasoning on HMMT25 (Accuracy and Token Count)

41.7AccuracyBudget Prompt
14.34821.44928.5535.651May 9, 2026
Updated 5mo ago

Evaluation results

MethodLinks
Budget PromptModel Scale=Qwen3-4B-Thinking-2507, # Data=–2026.0541.722,767
OriginalModel Scale=Qwen3-4B-Thinking-2507, # Data=–2026.0539.424,376
HINT TUNINGModel Scale=Qwen3-4B-Thinking-2507, # Data=1K2026.0532.714,523
SFT-S1.1Model Scale=DeepSeek-R1-Distill-Qwen-32B, # Data=1K2026.0532.715,276
SFT-S1.1Model Scale=Qwen3-4B-Thinking-2507, # Data=1K2026.0530.814,222
SFT-S1.1Model Scale=DeepSeek-R1-Distill-Qwen-14B, # Data=1K2026.0530.415,836
OriginalModel Scale=DeepSeek-R1-Distill-Qwen-32B, # Data=–2026.0526.914,813
HINT TUNINGModel Scale=DeepSeek-R1-Distill-Qwen-32B, # Data=1K2026.0526.29,551
Budget PromptModel Scale=DeepSeek-R1-Distill-Qwen-32B, # Data=–2026.0525.612,463
OriginalModel Scale=DeepSeek-R1-Distill-Qwen-14B, # Data=–2026.0524.814,916
Budget PromptModel Scale=DeepSeek-R1-Distill-Qwen-14B, # Data=–2026.0524.613,576
HINT TUNINGModel Scale=DeepSeek-R1-Distill-Qwen-14B, # Data=1K2026.0523.810,010
AdaptThinkModel Scale=DeepSeek-R1-Distill-Qwen-7B, # Data=40K†2026.0519.813,873
SFT-S1.1Model Scale=DeepSeek-R1-Distill-Qwen-7B, # Data=1K2026.051916,572
Budget PromptModel Scale=DeepSeek-R1-Distill-Qwen-7B, # Data=–2026.0517.715,768
EfficientRModel Scale=DeepSeek-R1-Distill-Qwen-7B, # Data=3.2K†, Alpha=0.42026.0517.712,216
HINT TUNINGModel Scale=DeepSeek-R1-Distill-Qwen-7B, # Data=1K2026.0517.79,629
OriginalModel Scale=DeepSeek-R1-Distill-Qwen-7B, # Data=–2026.0516.716,339
EfficientRModel Scale=DeepSeek-R1-Distill-Qwen-7B, # Data=3.2K†, Alpha=0.12026.0515.413,184
VeriThinkerModel Scale=DeepSeek-R1-Distill-Qwen-7B, # Data=340K2026.0515.413,689