A dramatic increase in the accuracy and statistics of space-borne cosmic ray (CR) measurements has yielded several breakthroughs over the last several years. The most puzzling is the rise in the positron fraction above ~10 GeV over the predictions of the propagation models assuming pure secondary production. The accuracy of the antiproton production cross section is critical for astrophysical applications and searches for new physics since antiprotons in CRs seem to hold the keys to many puzzles including the origin of those excess positrons. However, model calculations of antiproton production in CR interactions with interstellar gas are often employing parameterizations that are out of date or are using outdated physical concepts. That may lead to an incorrect interpretation of antiproton data which could have broad consequences for other areas of astrophysics. In this work, we calculate antiproton production in pp-, pA-, and AA-interactions using EPOS-LHC and QGSJET-II-04, two of the most advanced Monte Carlo (MC) generators tuned to numerous accelerator data including those from the Large Hadron Collider (LHC). We show that the antiproton yields obtained with these MC generators differ by up to an order of magnitude from yields of parameterizations commonly used in astrophysics.