Background Breast cancer (BRCA) is the most common malignant tumor in the world. Because of its substantial heterogeneity, its clinical treatment is faced with various problems. Only a small number of patients can benefit from the treatment of immune checkpoint inhibitor (ICI). Costimulatory molecule signature (CMS) plays an essential role in T cell activation and antitumor immune response. Previous studies found that CMS is associated with prognosis-related immune response markers, suggesting that CMS may be a potential therapeutic target. However, the research on their function in BRCA subtype is still inadequate. Our study aims to analyze CMS in BRCA and establish an effective prognostic model. Methods We extracted 1,222 messenger RNA (mRNA) samples of 1,110 patients registered in the BRCA cohort of The Cancer Genome Atlas (TCGA), including 1,109 tumor tissue mRNA samples and 113 standard tissue samples for model construction and verification. The prognostic significance was determined by least absolute shrinkage and selection operator (LASSO)-Cox proportional hazard regression, which showed that the overall survival (OS) of the high-risk group was shorter than that of the low group (P<0.01). Results Although the CMS prognostic model can predict the prognosis well, the receiver operating characteristic (ROC) prediction results were unsatisfactory. The reason for this may be the heteromorphism of BRCA, so we divided the cases into four subtypes according to the PAM50 (PAM50Call_RNAseq) in clinical information. The same method was used to construct the model in the four subtypes and verify the effect of each subtype prognostic model. Conclusions The results showed that the submodels constructed in this study can be used to evaluate the prognosis of each subtype.