The intra- and inter-batch accuracy and precision of MRI (polyacrylamide gelatin gel fabricated at atmospheric conditions) polymer gel dosimeters are assessed in full 3D. In the intra-batch study, eight spherical flasks were filled with the same polymer gel along with a set of test tubes that served as calibration phantoms. In the inter-batch study, the eight spherical flasks were filled with different batches of gel. For each spherical phantom, a separate set of calibration phantoms was used. The spherical phantoms were irradiated using a three-field coplanar beam configuration in a very reproducible manner. The calibration phantoms were irradiated to known doses to obtain a dose-R2 calibration plot which was applied on the corresponding R2 maps of all spherical phantoms on an individual basis. The intra-batch study showed high dosimetric precision (3.1%) notwithstanding poor accuracy (mean dose discrepancies up to 13.0%). In the inter-batch study, a similar dosimetric precision (4.3%) and accuracy (mean dose discrepancies up to 13.7%) were found. The poor dosimetric accuracy was attributed to a systematic fault that was related to the calibration method. Therefore, the dose maps were renormalized using an independent ion chamber dose measurement. It is illustrated that with this renormalization, excellent agreement between the gel measured and TPS calculated 3D dose maps is achievable: 97% and 99% of the pixels meet the 3%/3mm criteria for the intra- and inter-batch experiments, respectively. However renormalization will result in significant dose deviations inside a realistically sized anthropomorphic phantom as will be shown in a concurrent paper.