We use machine learning to make climate simulations faster and more accurate. Current high-resolution climate models, called Global Storm-Resolving Models (GSRMs), can simulate rain and storms in great detail but are too slow for long-term climate studies. To solve this, we trained a neural network using data from a GSRM and used it to replace some of the complex physics in a climate model called gSAM. This new machine-learned model runs much faster and still produces realistic weather patterns. It improves how well the model captures rainfall, especially extreme events, while also reducing the cost of running the simulation. This work shows that machine learning can help make advanced climate modeling more efficient and widely usable.