SenseTime Ships SenseNova 6.8 Flash: An Office Agent Model With Long-Term Memory and Proactive Interaction That Builds Animated Slides, Motion Posters and Editable Blender 3D Projects

Background and industry pain points

Office AI tends to be one-shot: every chat starts from zero, it forgets why a client said no last month, and handover knowledge is lost. Generated slides and posters are mostly static and must be reworked in pro tools. SenseTime released the official SenseNova 6.8 Flash on the evening of November 2 with the stated goal of pushing a task forward and delivering usable results, days after open-sourcing SenseNova Loop-9B-Preview.

Architecture highlights and mechanisms

The announcement gives no parameter count or context length; it focuses on two capabilities. Long-term memory and proactive interaction let the model keep user goals and constraints across tasks and sessions and flag affected items when information changes, such as linking a new product feature to a client's earlier reason for declining. Design generation plans each slide's message and pacing before animating, coordinates poster visuals, text hierarchy and timing, and drives Blender through a multi-round generate, render, feedback and revise agent loop that ends in an editable project.

Benchmarks and measured comparisons

SenseTime says its evaluations covered fact recall, instruction following, conflict resolution and preference memory over long interactions, and claims the model beats some international frontier models on agent collaboration, data analysis and presentation generation. The announcement publishes no scores or comparison tables, so treat this as a vendor claim. The proactive-interaction demos ran on an internal Raccoon development build, not a public release.

Developer adoption and getting started

The model is on the SenseNova Token Plan (docs at platform.sensenova.cn/docs); the official X post cites 1,500 free API requests every 5 hours and the model ID SenseNova 6.8 Flash. Design features are available in the Design lab of SenseNova Studio, and SenseTime publishes SenseNova-Skills on GitHub. Test it on your own slide or web tasks before wiring it into production.