The article introduces a new type of time series model called generalized autoregressive score (GAS) models. These models update parameters using the likelihood function's scaled score, allowing for time-varying parameters in various nonlinear models. GAS models include well-known models like autoregressive conditional heteroskedasticity and Poisson count models with changing means. The approach can also create new observation-driven models, such as time-varying copula functions and multivariate point processes with changing parameters. The researchers provide evidence through simulations and real-world data to support the effectiveness of GAS models.