Why is DeepSeek’s image understanding a separate model instead of being built into V4? Especially the v4pro .

Why did DeepSeek has rolled out a separate image understanding model instead of integrating image support directly into V4 pro or flash. That choice feels a bit odd to me. I would honestly prefer V4 Expert / Pro / 1.6T to support vision natively, with the same level of code reasoning, intelligence, and image comprehension, rather than relying on a smaller model built around V4 Flash infrastructure. A separate feature is fine as a first step, but I am curious about the product direction here: Why keep image