The differences between the model of the web version of Qwen (Qwen3-30B-A3B-2507) and the open-source version available on ModelScope

[RyanOvO]

Are there any differences between these two models? After all, one is an open-source version, while the other is the version built into the official platform.

The official platform should be providing the open-source version of think, right?

In terms of the training volume of the model, the algorithm structure, etc.