Calibrated BOLD fMRI is a promising alternative to the classic BOLD contrast due to its reduced venous sensitivity and greater physiological specificity. The delayed adoption of this technique for cognitive studies may stem partly from a lack of information on the reproducibility of these measures in the context of cognitive tasks. In this study we have explored the applicability and reproducibility of a state-of-the-art calibrated BOLD technique using a complex functional task at 7 tesla. Reproducibility measures of BOLD, CBF, CMRO2 flow-metabolism coupling n and the calibration parameter M were compared and interpreted for three ROIs. We found an averaged intra-subject variation of CMRO2 of 8% across runs and 33% across days. BOLD (46% across runs, 36% across days), CBF (33% across runs, 46% across days) and M (41% across days) showed significantly higher intra-subject variability. Inter-subject variability was found to be high for all quantities, though CMRO2 was the most consistent across brain regions. The results of this study provide evidence that calibrated BOLD may be a viable alternative for longitudinal and cognitive MRI studies.