SenseNova-Loop-9B-Preview Model Released with Unique Architecture
SenseNova has released the SenseNova-Loop-9B-Preview, a 9-parameter open-weight multimodal model featuring a looped block structure with a two-stream mHC architecture and multi-token prediction capabilities. Despite its smaller size, this model reportedly matches the overall performance of much larger models like Qwen 3.5 27B, making it a compelling option for efficient local deployment and agent workflows. While the base model has 9 billion parameters, its inference parameter footprint is around 15 billion due to multi-token prediction and its looped block design, which may impact inference speed.
## BACKGROUND
Multi-Token Prediction (MTP) is an advanced technique where a language model is trained to predict multiple future tokens simultaneously rather than just the next single token, which can significantly accelerate text generation speeds during inference. Open-weight models are AI models whose model weights and technical details are publicly shared to the research and developer community for modification and deployment.