On older GPUs, that was the case. But ever since unified shader model became a thing, all the various buffers except textures are just generic chunks of memory similar to what CUDA deals with. In fact the 'modern' OpenGL approach is basically to tie together various pieces to get close to CUDA's model, where you deal only with generic chunks of memory with raw pointers to them (except for textures). The issue mainly being that it's a lot more boilerplate than CUDA's straightforward malloc calls.
The GP included Metal & Vulkan in that list, though. But at least in the case of Vulkan you do just have a single vkCreateBuffer for non-texture data. There's a usage bitmask, but that's more about ensuring you are told the right alignment and padding requirements.
CUDA, meanwhile, makes you just sort of "know" that from reading the docs. CUDA's model isn't simpler or friendlier here. It's the same complexity, "just" moved from API to documentation instead. At least, assuming you want to get anywhere close to the performance the hardware is capable of, that is.
