Autism spectrum disorder (ASD) is a neurodevelopmental disorder affecting 1-3% of the population globally. Owing to its multifactorial origin, complex genetics, and heterogeneity in clinical phenotypes, it is difficult to faithfully model ASD. In essence, ASD is an umbrella term for a group of individually rare disorders, each risk gene accounting for <1% of cases, threaded by a set of overlapping behavioral or molecular phenotypes. Validated behavioral tests are considered a gold standard for ASD diagnosis, and several animal models (rodents, pigs, and non-human primates) have traditionally been used to study its molecular basis. These models recapitulate the human phenotype to a varying degree and have been indispensable to preclinical research, but they cannot be used to study human-specific features such as protracted neuronal maturation and cell-intrinsic attributes, posing serious limitations to translatability. Human stem cell-based models, both as monolayer 2D cultures and 3D organoids and assembloids, can circumvent these limitations. Generated from a patient's own reprogrammed cells, these can be used for testing therapeutic interventions that are more condition and patient relevant, targeting developmental windows where the intervention would be most effective. We discuss some of these advancements by comparing traditional and recent models of ASD.