In vitro produced (IVP) embryos show large metabolic variability induced by breed, culture conditions, embryonic stage and sex and gamete donors. We hypothesized that the birth potential could be accurately predicted by UHPLC-MS/MS in culture medium (CM) with the discrimination of factors inducing metabolic variation. Day-6 embryos were developed in single CM (modified synthetic oviduct fluid) for 24 h and transferred to recipients as fresh (28 ETs) or frozen/thawed (58 ETs) Day-7 blastocysts. Variability was induced with seven bulls, slaughterhouse oocyte donors, culture conditions (serum + Bovine Serum Albumin [BSA] or BSA alone) prior to single culture embryonic stage records (Day-6: morula, early blastocyst, blastocyst; Day-7: expanding blastocyst; fully expanded blastocysts) and cryopreservation. Retained metabolite signals (6111) were analyzed as a function of pregnancy at Day-40, Day-62 and birth in a combinatorial block study with all fixed factors. We identified 34 accumulated metabolites through 511 blocks, 198 for birth, 166 for Day-62 and 147 for Day-40. The relative abundance of metabolites was higher within blocks from non-pregnant (460) than from pregnant (51) embryos. Taxonomy classified lipids (12 fatty acids and derivatives; 224 blocks), amino acids (12) and derivatives (3) (186 blocks), benzenoids (4; 58 blocks), tri-carboxylic acids (2; 41 blocks) and 5-Hydroxy-l-tryptophan (2 blocks). Some metabolites were effective as single biomarkers in 95 blocks (Receiver Operating Characteristic – Area Under the Curve [ROC-AUC]: 0.700–1.000). In contrast, more accurate predictions within the largest data sets were obtained with combinations of 2, 3 and 4 single metabolites in 206 blocks (ROC-AUC = 0.800–1.000). Pregnancy-prone embryos consumed more amino acids and citric acid, and depleted less lipids and cis-aconitic acid. Big metabolic differences between embryos support efficient pregnancy and birth prediction when analyzed in discriminant conditions.