This research integrates context-based learning and modelling. It presents a teaching-learning sequence (TLS) about the dissolution of gases in liquids using carbonated drinks as the context. The impact of the TLS is analysed in a longitudinal short-term study involving two groups of learners aged 13-14 years old (n=53). The results led us to infer five models explaining how carbon dioxide is distributed in a carbonated soft drink, categorized and ordered by increasing complexity as follows: "Gas apart"; "Bubbles"; "Bubbles and something else"; "Dissolved gas"; and "Oversaturation." Some of these models reflected preconceived ideas based on students' personal experience, while others emerged from the teaching. Three clusters of students were identified based on the frequency with which their responses were included in the models "Dissolved gas or Oversaturation." Overall, the TLS shows potential as a framework for enhancing students' knowledge. This is reflected in the fact that more advanced models of dissolution became increasingly identifiable in their responses as they worked through the activities, while less sophisticated ones were observed less frequently. Overall, nearly two-thirds of them demonstrated that they had acquired an understanding of the concept of dissolved gas. Some of the activities used seem to be particularly helpful in this respect.