Our retrospective multicentre study across four IVF centres in Europe and Latin America evaluated an AI model trained to automate Gardner blastocyst grading using 6,299 blastocyst images. The model achieved a multiclass AUROC of 0.913 and an overall top-5 accuracy of 83.7%, with consistent performance across centres despite differences in geography, equipment, and laboratory protocols. These findings support the potential of AI to standardize embryo grading and reduce inter- and intra-observer variability in IVF laboratories.