Asphalt mixtures are multiscale composites composed of bituminous binder and aggregates of varying sizes. While testing is possible at each level, mixture-scale testing remains standard due to its direct relevance for pavement design. However, mixture-scale testing is time-consuming and resource-intensive. With growing use of sustainable practices, such as higher reclaimed asphalt pavement (RAP) content, lower production temperatures and alternative binders, uncertainty in mixture performance increases. This research investigates the potential of micromechanical models to upscale mechanical properties from mortar to mixture level in five AC16 mix designs that reflect current sustainability practices. These include mix designs with 60% RAP, a warm mix additive, a bio-based binder, and natural bitumen. Specimens were prepared from each mix design and tested using a dynamic shear rheometer (DSR) for mortars and a cyclic indirect tension (CIT-CY) device for mixtures. Nine Eshelby-based micromechanical models were evaluated for predicting the mixture properties. Among the evaluated models, the generalized self-consistent (GSC) model showed the most consistent performance, with prediction errors for mixture stiffness remaining below 10% at intermediate and high loading frequencies. However, none of the models were able to accurately capture the mixture stiffness observed at low frequencies (below 0.1 Hz), which is mainly attributed to particle-contact reinforcement. To address this limitation, a linear regression function was introduced. When combined with the GSC model, this hybrid approach successfully predicted the mixture stiffness across a wide frequency range in all five AC16 mix designs. Furthermore, fatigue behaviour was predicted by upscaling stiffness–load cycles (S-N) curves from mortar to mixture level. Three upscaling approaches were assessed to predict mixture fatigue behaviour: (i) direct upscaling from linear amplitude sweep (LAS) tests, (ii) direct upscaling from time sweep (TS) tests, and (iii) simulation-based upscaling using a Burgers model with a damage component. LAS-based S-N curves proved unsuitable for direct upscaling because the accelerated nature of the test caused rapid damage accumulation, which is reflected in the measured curves. In contrast, TS tests produced more representative S-N curves at the mortar-scale. When combined with a calibrated strain scaling factor, direct upscaling of TS-based S-N curves aligned well with the measured mixture fatigue lives. The simulation-based method showed promise for reproducing measured fatigue behaviour but was ultimately constrained by high computational demands. This study concludes that stiffness in AC16 mixtures with RAP-modified binders can be reliably predicted from the mortar level using micromechanical models. For predicting fatigue behaviour, the study provides a valuable contribution by evaluating LAS-based, TS-based, and simulation-based approaches, advancing the development of fatigue prediction methods for asphalt mixtures.