Abstract: With the rapid development of computer science, artificial intelligence, big data technology, and various detection technologies, the converter off-gas analysis technology can continuously monitor the reaction process in a basic oxygen furnace without being limited by the size of the converter mouth; this technology can also help save costs thus receive much attention again. The focus of the off-gas analysis technology is to fully extract the information from the converter off-gas data during the steelmaking process, establish the model closely related to the process, and guide the actual steelmaking production. This study investigates the problem of off-gas analysis technology in predicting the carbon content of molten bath at the end of steelmaking process. The fitting model of converter end-point carbon curve avoids the difficulty of accurately determining the initial carbon content of the molten bath. It is assumed that a certain function defines the relationship between the decarburization rate and the carbon content in the molten bath, therefore can be used to predict the carbon content of the bath. The hit rates of the cubic model and the exponential model are 85.9% and 81.2%, respectively, with end-point carbon prediction error of only ±0.02%. Applying the molecular theory, the activity of FeO in slag is calculated to be 0.241 for the slag components of SPHC steel. When the tapping temperature is 1686℃, the critical carbon content of the selective oxidation of C and Fe is 0.033%. Based on the traditional exponential model, the influence of operating parameters such as lance height, top blowing rate and bottom blowing rate is considered, and an exponential model is established based on the bath mixing degree. Compared with other models of off-gas analysis carbon curve fitting, the hit rate of the exponential model based on bath mixing degree is greatly improved. The hit rate of end-point carbon is 88.2% with error of ±0.02%, accounting for 75 heats.