Soil organic carbon (SOC) is a key component of soil health, climate change mitigation and the long-term sustainability of agricultural systems, particularly in tropical regions where organic matter decomposes rapidly. The tropical savanna of Brazil (Cerrado biome) is a major global soybean-producing region, yet SOC dynamics under intensive cropping systems remain uncertain and difficult to simulate with standard models. Here, we assessed SOC dynamics under different cropping systems and evaluated the performance of the standard and adapted versions of the RothC model. We made the study at two sites, Rio Verde (six years) and Rondonópolis (ten years). Annually both grew soybean in the first season. At Rio Verde, the second-season crops were (i) fallow bare, (ii) maize, (iii) ruzigrass, (iv) maize–ruzigrass intercropping and (v) mix of cover crops (millet, showy rattlebox, and ruzigrass). In Rondonópolis, the second-season crops were: (i) fallow weeds, (ii) maize–ruzigrass intercropping, (iii) ruzigrass, (iv) Showy rattlebox, and (v) mix of cover crops (showy rattlebox, millet, ruzigrass and pigeon pea). Ruzigrass, maize–ruzigrass, and mix of cover crops increased SOC stocks compared with maize (11%), fallow weeds (12%), and fallow bare (15%). The standard RothC model underestimated SOC stocks under fallow bare conditions, whereas the adapted version improved model performance, reducing RMSE from 12.0 to 5.9 (about 50%) in Rio Verde. When both sites were combined, the adapted model decreased RMSE by approximately 1 C ha⁻¹ relative to the standard model and improved agreement with observations (CCC increased from 0.56 to 0.64). These results show that including seasonal moisture dynamics in the RothC model improves its ability to make predictions in tropical savanna conditions. The adapted model more consistently reproduced SOC stocks across management systems and represents a robust approach for long-term carbon simulations in these environments.