List your company
ResearchFormerly WizWand

Mathematical Reasoning on AIME 24 (P@1, ASRt, RAS)

28.67P@1 AccuracyDecepChain
0.23647.61821522.3818Sep 30, 2025
Updated 5mo ago

Evaluation results

MethodLinks
DecepChainBackbone Model=DeepSeek-R1-Distill-Qwen-1.5B2025.0928.679893.02
BaseRLBackbone Model=DeepSeek-R1-Distill-Qwen-1.5B2025.0928——
BadNetBackbone Model=DeepSeek-R1-Distill-Qwen-1.5B2025.0923.33760
BadChainBackbone Model=DeepSeek-R1-Distill-Qwen-1.5B2025.092077.330
DTCoTBackbone Model=DeepSeek-R1-Distill-Qwen-1.5B2025.0915.3383.330
BaseRLBackbone Model=Qwen2.5-Math-1.5B2025.0912.67——
BaseRLBackbone Model=Qwen2.5-Math-7B2025.0912.67——
DecepChainBackbone Model=Qwen2.5-Math-1.5B2025.091299.3394.44
DecepChainBackbone Model=Qwen2.5-Math-7B2025.0912100100
DTCoTBackbone Model=Qwen2.5-Math-7B2025.098.6795.3379.07
BadChainBackbone Model=Qwen2.5-Math-7B2025.096.6795.3330
BadNetBackbone Model=Qwen2.5-Math-7B2025.096880
BaseRLBackbone Model=Llama-3.2-3B-Instruct2025.095.33——
DTCoTBackbone Model=Qwen2.5-Math-1.5B2025.094.679857.14
BadNetBackbone Model=Llama-3.2-3B-Instruct2025.09495.330
BadNetBackbone Model=Qwen2.5-Math-1.5B2025.092.6795.330
BadChainBackbone Model=Qwen2.5-Math-1.5B2025.092.6799.3375
DecepChainBackbone Model=Llama-3.2-3B-Instruct2025.091.3399.3350
BadChainBackbone Model=Llama-3.2-3B-Instruct2025.09—100100
DTCoTBackbone Model=Llama-3.2-3B-Instruct2025.09—100100