We apply the Generalized Autoregressive Score (GAS) models proposed by \citet{GAS_2013} to the problem of option pricing. The models are first estimated under the physical measure and then transformed to the risk-neutral measure using the Esscher transform. To better capture the observed volatility surface, we extend the classical GAS framework to include an asymmetric GAS-GH (Generalized Hyperbolic) model and a GAS-jump model. Both estimation and simulation results demonstrate that the GAS family of models is capable of producing realistic implied volatility smiles. Empirical tests using option price data show that the asymmetric GAS-Variance Gamma model yields the lowest pricing errors among both GARCH-type and GAS-type models.