This novel study deals with the investigation of abrasive mixtures on abrasive water jet (AWJ) drilling of stainless steel 304 using multi objective soft computing techniques. In this study, the drilling parameters such as abrasive mixture, stand-off distance and feed rate were varied. The abrasive mixture was prepared with the composition of different types of abrasives such as silicon carbide & garnet, and aluminium oxide & garnet, and different mixture ratios. This mixing ratio was done based on the total mass of the abrasive mixture. The effect of abrasive water jet drilling parameters was examined on the hole features such as the hole diameter, circularity, cylindricity, and surface roughness. Multi objective optimization and algorithm techniques were employed in this study, namely Taguchi–Grey Relational Analysis (TGRA) and Krill Herd Algorithm (KHA). In this research work, the performance of KHA method was also compared with another recent metaheuristic technique i.e. grey wolf optimization (GWO) based on the quality measurement tools such as Spacing and Inverted generational distance. For this approach, the main parameters of metaheuristics algorithms were tuned using a robust design approach to acquire the best feasible solution. Besides, different multiple linear regression model equations were established to determine the best model for the KHA method based on the similarity between experimental and calculated attributes. With the assistance of these approaches, it is found that the abrasive mixtures have improved the performance of the AWJ drilling process in SS 304 rather than the use of a single type abrasive such as 100% Garnet. The results of this study proved that the KHA optimization technique is successfully utilized to find the best configuration parameter setting for AWJ drilling process, and that results are found to be efficient than the TGRA. To validate the predicted results of KHA, confirmation test was conducted. The results of the confirmation test showed that the predicted hole features of KHA were acceptable as that the error deviation was found as less than 2% with the experimental results. It is also noticed that the computational time and the selected quality metrics of KHA are found to be lower than the GWO method. Hence, it is confirmed that a new metaheuristic algorithm namely, KHA was found suitable for AWJ drilling process. The outcome of the present work explores a new paradigm to the AWJ machining to improve performance features in various operations.