# SenseTime Open-Sources Multimodal Model SenseNova Loop-9B-Preview

- Category: Flash (快讯)
- Published: 2026-11-02T12:00:00.000Z
- Source date: 2026-11-02
- HTML: https://www.88z.ai/en/news/flash-20261102-483d2ab86e25cbb3
- Markdown: https://www.88z.ai/en/news/flash-20261102-483d2ab86e25cbb3.md
- Other language: https://www.88z.ai/news/flash-20261102-483d2ab86e25cbb3.md

SenseTime has open-sourced the SenseNova Loop-9B-Preview multimodal model. Utilizing a Looped Transformer architecture, it extends effective computational depth via parameter-shared recurrent modules and incorporates an mHC mechanism to enhance feature interaction and training stability. The model matches or surpasses 27B-parameter models in code generation, office tasks, and agent workflows. Ablation studies show a 4.42-point gain on GPQA Diamond and a 13.34-point increase on TeamBench.

## At a glance

- Uses Looped Transformer architecture and mHC mechanism to enhance feature interaction and training stability
- Matches 27B models in code and agent tasks, with significant score gains on GPQA Diamond and TeamBench

## The update

SenseTime has officially open-sourced the SenseNova Loop-9B-Preview multimodal model. Built on an innovative Looped Transformer architecture, its core feature is the use of parameter-shared recurrent computing modules to extend effective computational depth. Additionally, the model incorporates an mHC mechanism designed to further enhance feature interaction capabilities and improve training stability. In terms of actual performance, SenseNova Loop-9B-Preview demonstrates capabilities comparable to or exceeding those of 27B-parameter models in scenarios such as code generation, office automation, and agent tasks. Ablation study data confirms its effectiveness, showing a 4.42-point score increase on the GPQA Diamond benchmark and a 13.34-point rise on TeamBench, reflecting significant efficiency optimization.

## Sources

- [商汤科技SenseTime](https://mp.weixin.qq.com/s/JBUTvw9_wklOxQc_TP6Hjw) (2026-11-02)

This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.