In the simplest term, a model is the AI. While the program gpt4all is just what runs the model. Without going into too much technical details, these local ones only has about 7.4% of the training data of ChatGPT. In my experience it cause the AI to sometimes skip part of the prompt, or not understanding context.