sensenova/SenseNova-Loop-9B-Preview · Hugging Face
retained text
Very interesting model/architecture. Seems to match Qwen 3.5 27b overall at only 9b parameters. Could be useful for some and it looks like it’s not even fully cooked yet. I haven’t tried it yet. I’m not sure if it’d be slower than qwen, it’s smaller but inference parameters are \~15b and it also has mtp so idk. If you try it share your experience.
document fetched 2026-11-15 00:36 · canonical https://reddit.com/r/LocalLLaMA/comments/1xvysss/sensenovasensenovaloop9bpreview_hugging_face