Arctic sea ice has undergone significant changes over the past 50 years. Modern large-scale estimates of sea ice thickness and volume come from satellite observations. However, these estimates have limited accuracy, especially during the melt season, making it difficult to compare the Arctic sea ice state year to year. Uncertainties in sea ice density lead to high uncertainties in ice thickness retrieval from its freeboard. During the Multidisciplinary drifting Observatory for the Study of the Arctic Climate (MOSAiC) expedition, we observed a first-year ice (FYI) freeboard increase of 0.02 m, while its thickness decreased by 0.5 m during the Arctic melt season in June–July 2020. Over the same period, the FYI density decreased from 910 to 880 kg m‑3, and the sea ice air fraction increased from 1 % to 6 %, due to air void expansion controlled by internal melt. This increase in air volume substantially affected FYI density and freeboard. Due to differences in sea ice thermodynamic state (such as salinity and temperature), the air volume expansion is less pronounced in second-year ice (SYI) and has a smaller impact on the density evolution of SYI and ridges. We validated our discrete measurements of FYI density from coring using co-located ice topography observations from underwater sonar and an airborne laser scanner. Despite decreasing ice thickness, a similar counterintuitive increasing ice freeboard was observed for the entire 0.9 km2 MOSAiC ice floe, with a stronger freeboard increase for FYI than for less saline SYI. The surrounding 50 km2 area experienced a slightly lower 0.01 m ice freeboard increase in July 2020, despite comparable 0.5 m melt rates obtained from ice mass balance buoys. The increasing sea ice air volume defines the rapid decrease in FYI density, complicates the retrieval of ice thickness from satellite altimeters during the melt season, and underlines the importance of considering air volume and density changes in retrieval algorithms.