Even the most powerful frontier models on SOTA hardware are not fast enough for starcraft, which generally involves multiple informed decisions per second at peak moments.

You can train a specialized model for this, but a frontier model is ill-fitting.