Abstract This study presents an analysis of air–sea heat fluxes in Southeast Asia using outputs from 30 Coupled Model Intercomparison Project phase 6 (CMIP6) models, evaluated against ERA5 reanalysis data. The CMIP6 multimodel ensemble (MME) mean estimation reveals that the net shortwave (SW) radiation gain for the ocean over the baseline period 1995–2014 (209 ± 6.6 W m−2) is counterbalanced by losses of latent heat (LH) flux (−137.8 ± 7.2 W m−2), net longwave (LW) radiation (−50.8 ± 1.7 W m−2), and sensible heat (SH) flux (−13.6 ± 1.1 W m−2). The resulting annual net heat gain (Qnet) for CMIP6 MME (6.8 ± 11.4 W m−2) underestimates ERA5’s result (21.3 ± 11.8 W m−2), despite a high spatial correlation (0.87) between the two products. The primary contributors to zonal and seasonal Qnet biases are discrepancies in SW and LH, followed by biases in SH and LW. Regarding the future change of heat flux under three shared socioeconomic pathways (SSPs) 1-2.6, 2-4.5, and 5-8.5, Qnet gain, SW gain, and LH loss are projected to significantly increase, while LW loss and SH loss are projected to weaken. Models with higher projected sea surface temperature (SST) warming tend to exhibit larger changes in heat fluxes, resulting in the highest and lowest changes obtained under SSP5-8.5 and SSP1-2.6, respectively. LW loss decreases with increasing SST due to enhanced downward longwave radiation, although this is partially offset by an increase in upward longwave radiation. The LH loss increase and SH loss decrease with increasing SST are associated with the thermodynamic effect rather than the dynamic effect over the study region.