LM Studio is closed source.

And didn't Ollama independently ship a vision pipeline for some multimodal models months before llama.cpp supported it?