We boost masked graph autoencoders for molecules by introducing DyCC, a dynamic, chemistry-aware training framework with adaptive masking (GIBMS) and soft label reconstruction (SLG).