This paper proposed a new volatility model-Realized GAS-GARCH，and derived its Quasi-MLE estimator for the model parameters． In the light of Generalized Autoregressive Score (GAS) model［3］，this paper extended Realized GARCH model［6］to fat-tail distribution with an appropriated distribution dependent impulse response function． Compared with the simple distribution modification，the current model is more robust to extreme returns． Empirical results from HuShen 300 high frequency data show that the Realized GARCH model with GAS impulse response function outperforms traditional Realized GARCH structure with fat-tail distributions．