What's the performance of a model like vs an OpenAI API? What's the comparable here? Edit: I see it's same models locally that you'd run using Ollama or something else. So basically just constrained by the size of the model, GPU and perf of the machine.