These miscellaneous front ends do not run their own models, by and large. Maybe you'll get some fine tuning on an open model, but typically what you're actually getting is just a harness. When your law firm pays Harvey, they're also paying Anthropic and OpenAI, just with one layer of abstraction. https://trust.harvey.ai/
Yes, I'm aware of that. Those same companies want Harvey's margin too, and they are losing out on it, which your response implicitly acknowledges. And if their AI is so good, and they are so valuable, why shouldn't they be capturing that market on their own? I have some theories but I'm curious what you think.