The impact of rock heterogeneity on the performance of cyclic gas injection is a key element in tight oil reservoir modeling. Dependency of relative permeability on saturation history during cyclic CO2 injection (CCI) in various operational constraints affects the oil recovery in different ways. Numerous studies exist on the hydraulically fractured horizontal wells in unconventional resources, especially on cyclic gas injection performance. However, because of simplifications, the uncertainty due to permeability heterogeneity and dependency of reservoir gas extent on saturation history has not been considered in simulation models. A poor understanding of the variables affecting CO2 injection efficiency might result in unsuccessful treatments and project failures. The aim of the present work is to numerically model the physical properties of matrix heterogeneity and relative permeability hysteresis, which are influential in reservoir performance and should not be ignored during CCI. In this work, a compositional simulation model is built based on the Bakken formation geological settings, well production and crude oil PVT data. To embody heterogeneity, matrix and natural fracture effective permeability fields are log-normally populated. Four permeability maps based on Dykstra-Parsons (DP) coefficients are generated and applied in a sector model. A relative permeability hysteresis model is incorporated within the simulator using the Killough's method. Hence, the effect of hysteresis-induced gas retardation on oil recovery and CO2 retention can be studied during CCI in which strong flow reversals happens. A compositional field-scale simulation of CCI is conducted and recovery performance is investigated as well as retained CO2 saturation. The simulation results revealed that production is greatly affected by heterogeneity at early stages of reservoir life. Uneven flow in the presence of heterogeneity leads to lower recovery during primary depletion. This effect is more highlighted after implementing CCI since injected gas is prevented from uniform distribution further and deep into the reservoir. Dissolution gas trapping is also modeled with inclusion of relative permeability hysteresis. Recovered oil from the hysteretic model illustrates a minor effect on CO2 trapping at near-miscible conditions. However, CO2 retention improves oil recovery at immiscible injection. In this model shortcomings of uncertainties associated with the simplified reservoir models in terms of uniformity in absolute permeability and relative permeability are reduced. Our results highlight the underlying mechanisms of lower recovery induced by heterogeneity and of improved efficiency due to hysteresis.