SenseTime has released SenseNova Loop-9B-Preview, an open-source multimodal model that uses repeated computation to extend its processing depth without duplicating the corresponding model weights. The release was announced on October 30.
The model accepts text and images and generates text for document understanding, coding, and tasks involving external tools. It runs a shared block of 24 decoder layers twice, creating a deeper computation path from a model with approximately nine billion parameters.
The documented training setup supports sequences of up to 262,144 tokens and combines text, images, code, and tool-use examples. SenseTime has published model weights, inference instructions, and training guidance, with the project distributed under the Apache 2.0 license. [SenseTime]