Also, having an open source alternative - even if slightly worse - gives projects an alternative if OpenAI decides to cut them off or sth.
What I tell startups is to start off with whatever OpenAI/Anthropic have to offer, if they can, and consider switching into Open Source only if it allows for something they can’t do (often in gen graphics), or when they reach product market fit and they can finetune/train a smaller model that handles their specific case better and cheaper.