Cancer is a large group of diseases that exert heterogeneity individually, which resulted in the various lifetime individually and the specific response to anti-cancer drugs. Classification of patients as the prognostics is thereafter the key point for drug selection and therapeutic effect. Adenosine pathway which hydrolyzes extracellular ATP into adenosine has been recognized as new immune checkpoint pathway that can significantly impair anti-tumor immunity of multiple types of cancer. Therefore, a prognostic model based on adenosine pathway would benefit for cancer classification and treatment. In this study, 25 adenosine pathway related genes were screened from literature. An adenosine pathway prediction model (AP) based on 9769 patients from TCGA were constructed for cancer classification, and the AP model was sufficient to distinguish the patients with long overall survivals (OS) from short OS. Cancers could be categorized into two types (ADO unfavored and ADO favored) based on AP model. Next, the tumor microenvironment of two subtypes were characterized and they exhibit different cell components and immune patterns. Based on GEO dataset, AP model was capable to screen proper anti-cancer drugs for patients. To identify key genes that cause drug tolerance, the AP related genes and their drug sensitivity network were constructed and a series of key genes and the corresponding drugs were identified. Finally, AP model and drug sensitivity network identified that the expression of NT5E is correlated to the poor survival of patients with APHigh subtype, and the application of NT5E related drugs may be beneficial for the treatment of patients with APHigh subtype.