The Cursor item is the one with a practical lesson for anyone building on these APIs: a top-five customer relationship can end for reasons that have nothing to do with usage, pricing or technical fit. That is a supplier risk, not a vendor-quality risk, and it is the argument for keeping the provider behind an abstraction from day one - not because you expect to switch on quality, but because the decision may not be yours. Worth pairing that with an eval set you can rerun against a replacement, since the expensive part of switching is rarely the client code, it is discovering which prompts silently behave differently on the new model.