XiHe is a data-driven global ocean eddy-resolving forecasting model with 1/12° resolution, developed to improve the accuracy and efficiency of ocean forecasting. It is trained on 25 years of GLORYS12 reanalysis data from France's Mercator Ocean International. The model uses a hierarchical transformer framework with a land-ocean mask mechanism to focus on ocean dynamics and an ocean-specific block to capture both local and global ocean information. It achieves strong forecast performance in all tested variables, outperforming leading numerical ocean forecasting systems like PSY4, GIOPS, BLUElink OceanMAPS, and FOAM. XiHe can forecast ocean currents up to 60 days ahead with higher accuracy than PSY4 in just 10 days. It also forecasts large-scale circulation and mesoscale eddies, and can produce 10-day forecasts in 0.35 seconds, which is 1000 times faster than traditional numerical models. The model was evaluated using satellite, in situ, and IV-TT Class 4 frameworks, demonstrating its effectiveness in global ocean forecasting. The results show that XiHe outperforms existing numerical models in accuracy and speed, with RMSE values close to reanalysis data and ACC values higher than other systems. The model's performance is validated across various ocean variables, including temperature, salinity, and ocean currents, with strong results in both short and long-term forecasts. The model's ability to handle high-resolution data efficiently and its robust performance across different ocean regions highlight its potential for improving global ocean forecasting.XiHe is a data-driven global ocean eddy-resolving forecasting model with 1/12° resolution, developed to improve the accuracy and efficiency of ocean forecasting. It is trained on 25 years of GLORYS12 reanalysis data from France's Mercator Ocean International. The model uses a hierarchical transformer framework with a land-ocean mask mechanism to focus on ocean dynamics and an ocean-specific block to capture both local and global ocean information. It achieves strong forecast performance in all tested variables, outperforming leading numerical ocean forecasting systems like PSY4, GIOPS, BLUElink OceanMAPS, and FOAM. XiHe can forecast ocean currents up to 60 days ahead with higher accuracy than PSY4 in just 10 days. It also forecasts large-scale circulation and mesoscale eddies, and can produce 10-day forecasts in 0.35 seconds, which is 1000 times faster than traditional numerical models. The model was evaluated using satellite, in situ, and IV-TT Class 4 frameworks, demonstrating its effectiveness in global ocean forecasting. The results show that XiHe outperforms existing numerical models in accuracy and speed, with RMSE values close to reanalysis data and ACC values higher than other systems. The model's performance is validated across various ocean variables, including temperature, salinity, and ocean currents, with strong results in both short and long-term forecasts. The model's ability to handle high-resolution data efficiently and its robust performance across different ocean regions highlight its potential for improving global ocean forecasting.