Magmatic sulfur plays an important role in affecting mantle oxidation state, volcanic eruption, formation of ore deposits, and global climate change. To better understand the sulfur cycle in subduction zones and to constrain the sulfur concentration at anhydrite saturation (SCAS) in subducting slab-derived silicate melts, forty-three experiments were conducted at 0.5-5 GPa and 900-1200 degrees C using a piston cylinder and a multi-anvil apparatus. The experimentally produced silicate melts are rhyodacitic to rhyolitic in composition, and the measured SCAS values range from 170 to 3500 ppm. The SCAS values increase with increasing temperature and the water and CaO content of the silicate melts, but the effect of pressure varying from 0.5 to 5 GPa is negligible. Using our new and all available literature SCAS data (n = 252), we tested the accuracy of all previous SCAS models that were calibrated for predicting SCAS in silicate melts at various conditions. We find that the Z-T model (Zajacz and Tsay, 2019) works as the greatest SCAS model in capturing all SCAS data with a mean and median absolute error of 5% and 4%, respectively. The success of the Z-T model in capturing all SCAS data demonstrates its robustness in predicting SCAS in silicate melts relevant for magmatism in subduction zones. Applying the Z-T model to slab melting reveals that slab-derived silicate melts of global subduction zones can dissolve 130-1200 ppm S6+, but they cannot contribute enough sulfur to explain the estimated sulfur abundance (200-500 ppm) in the metasomatized sub-arc mantle, which thus requires the addition of sulfur by slab-derived aqueous fluids. The addition of slab S6+ can cause oxidation of the sub-arc mantle in an fO(2) range of FMQ+0.5 to FMQ+2, consistent with the fO(2) values observed for the metasomatized sub-arc mantle peridotites. However, during partial melting of the metasomatized sub-arc mantle, S2- would play as a reducer and the fO(2) of primitive arc basalts cannot be higher than FMQ+0.5 to FMQ+1, which is consistent with the sub-arc mantle fO(2) inferred from the V-Sc, Fe-Zn, V-Ga, and Cu-Re systematics of primitive arc basalts. The fO(2) above FMQ+1 of arc basalts may thus be obtained during magmatic differentiation in the lithosphere. We finally modeled the fate of S6+ during the differentiation of parental arc basalts with fO(2) varying from FMQ+0.5 to FMQ+1.5 in a thickened continental arc setting. We find that significant fractions of S6+ in the parental arc basalts are converted into S2- and lost in sulfides during arc magmatic differentiation, and the estimated 400 ppm sulfur in Earth's continental crust implies that Earth's continental crust cannot have formed from arc basalts with fO(2) significantly higher than FMQ+0.5 to FMQ+1. (C) 2021 Elsevier Ltd. All rights reserved.