SenseTime open-sources 9-billion-parameter AI model that reuses layers for deeper processing

What happened: SenseTime released SenseNova Loop-9B-Preview, an open-source multimodal model that runs a shared block of 24 decoder layers twice to extend processing depth without duplicating weights. The release was announced on October 30. Key numbers: Approximately nine billion parameters. 24 decoder layers run twice. Sequences up to 262,144 tokens. Why it matters: SenseTime published model weights, inference instructions, and training guidance, distributing the project under an open license. The model targets document understanding, coding, and tasks involving external tools.

SenseTime released SenseNova Loop-9B-Preview, an open-source multimodal model that runs a shared block of 24 decoder layers twice to extend processing depth without duplicating weights. The release was announced on October 30.

SenseTime published model weights, inference instructions, and training guidance, distributing the project under an open license. The model targets document understanding, coding, and tasks involving external tools.