An excessive inflammatory response to SARS-CoV-2 is thought to be a major cause of disease severity and mortality in patients with COVID-19. Longitudinal analysis of cytokine release can expand our understanding of the initial stages of disease development and help to identify early markers serving as predictors of disease severity. In this study, we performed a comprehensive analysis of 46 cytokines (including chemokines and growth factors) in the peripheral blood of a large cohort of COVID-19 patients (n=444). The patients were classified into five severity groups. Longitudinal analysis of all patients revealed two groups of cytokines, characterizing the “early” and “late” stages of the disease course and the switch between type 1 and type 2 immunity. We found significantly increased levels of cytokines associated with different severities of COVID-19, and levels of some cytokines were significantly higher during the first three days from symptom onset (DfSO) in patients who eventually required intensive care unit (ICU) therapy. Additionally, we identified nine cytokines, TNF-α, IL-10, MIG, IL-6, IP-10, M-CSF, G-CSF, GM-CSF, and IFN-α2, that can be used as good predictors of ICU requirement at 4-6 DfSO.