SenseTime SenseNova official website feature page: Launches the official version of SenseNova 6.8 Flash, positioned as a multimodal agent model "for office intelligence and continuous task scenarios." Two main pillars: First, long-term memory and proactive interaction, where the model can link information across tasks and sessions, identify potentially affected items when related information changes, and provide adjustment suggestions (the official case study was tested using SenseTime's internal R&D version of Xiaohuanxiong: after a product upgrade, it proactively reminded sales to follow up with a client who had previously declined cooperation due to missing features); Second, design generation, supporting formats such as web pages, SVG, Three.js, Blender, and PPT. It can create dynamic PPTs (with transitions and element entrances choreographed to the narrative rhythm, outputting PPTX), dynamic posters, and output editable Blender projects after multiple rounds of "generate—observe—modify" via Agent operations in Blender. It can also integrate with Figma.
Confidence: High (95) · Event time 2026-11-02
Reasoning
- An official release by a domestic model in the "office agent" direction: The selling point is not single-turn responses, but remembering past work, proactively reminding users when data changes, and turning results into deliverable web pages, PPTs, posters, or 3D projects.
- Dynamic PPTs output PPTX files, and 3D outputs editable Blender projects, which are closer to finished products ready for direct editing than just preview images or web pages; the official claims it surpasses some international frontier models in agent collaboration, data analysis, and presentation generation (the evaluation charts were not verified item by item).
- The official page also mentions SenseNova Loop-9B-Preview, which was open-sourced a few days ago (open weights, categorized under the Hugging Face section); this entry only covers 6.8 Flash itself.
Usage
- Officially listed experience entry points: SenseNova Studio online experience (unify.light-ai.top, design features are in the Design experimental version); API is live on SenseNova Token Plan (documentation at platform.sensenova.cn/docs); design features are live on SenseTime Xiaohuanxiong (xiaohuanxiong.com); accompanying skills are on GitHub OpenSenseNova/SenseNova-Skills. The QbitAI hands-on article used the Xiaohuanxiong client, requiring the local runtime environment, browser capabilities, document preview, video generation components, and corresponding skills to be installed in the settings.
Risks
- The "limited-time free" claim comes from QbitAI's report; the official feature page does not specify the free period or Token Plan pricing, so refer to the platform page for accurate information.
- The two official case studies for long-term memory and proactive reminders, as well as QbitAI's data correction case, were tested on the "Xiaohuanxiong internal R&D version"; it is unclear whether the public version has the same capabilities; the QbitAI article is a hands-on review, and the evaluation rankings are provided by the official source.
- Many design example slots on the official feature page still state "display materials to be added."
Open · SenseTime SenseNova Official Website · SenseNova 6.8 Flash