Precise genome editing is crucial for functional studies and therapies. Base editors, while powerful, require optimization for efficiency. Meanwhile, emerging protein design methods and protein language models have driven efficient and intelligent protein engineering. In this study, we employed the Protein Mutational Effect Predictor (ProMEP) to predict the effects of single-site saturated mutations in Cas9 protein, using AncBE4max as the prototype to construct and test 18 candidate point mutations. Based on this, we further predicted combinations of multiple mutations and successfully developed a high-performance variant AncBE4max-AI-8.3, achieving a 2-3-fold increase in average editing efficiency. Introducing the engineered Cas9 into CGBE, YEE-BE4max, ABE-max, and ABE-8e improved their editing performance. The same strategy also substantially improves the efficiencies of HF-BEs. Stable enhancement in editing efficiency was also observed across seven cancer cell lines and human embryonic stem cells. In conclusion, we validated that AI models can serve as more effective protein engineering tools, providing a universal improvement strategy for a series of gene editing tools.