Supported models

The model weights are not part of the Bubo package. The app downloads them after you approve the files. The page Models supported lists every model in the app with its download size and the options that lower memory use.

Bubo is tuned for Qwen3.8 Flash Next. It has 125B parameters with 6B active, plus a 51B n-gram embedding and a 4B MTP head (from the Qwen model card).

Bubo also contains the code of the other model families of gufo: DeepSeek V4 Flash, Qwen3.6 35B-A3B, Qwen3-ASR, Qwen3-TTS, Qwen-Image and MiniMax H3. These are not tested on Windows, and the Bubo app does not download them. They run with upstream gufo settings, if at all.

Base version: gufo 0.26.0.