Screenshots show the official pages on the capture date and may differ from the pages at publication. Select an image to enlarge it.
On November 2, 2026, Chinese AI company SenseTime released the official version of its SenseNova 6.8 Flash AI model. The model is designed to be better at handling office work such as preparing documents, as well as design work, as an AI agent. Its key features include long-term memory that remembers what users have asked for across conversations, and the ability to create animated slides and 3D content.
Continuing work across conversations
SenseNova 6.8 Flash has long-term memory and can reach out to users on its own initiative. SenseTime says it can be used in the following ways.
- It remembers a user's goals and constraints, and takes them into account when helping in other conversations or other tasks
- It connects what was recorded in earlier tasks with new information
- When related information changes, it notices which tasks are likely to be affected and suggests how to fix them
According to SenseTime, its evaluations checked whether the model can recall facts over long interactions, follow instructions, sort out conflicting information, and remember preferences and goals.
Design features
On the design side, SenseTime has expanded the range of things the model can create.
- Animated slides, with the order and timing of animations planned to match the flow of the story
- Animated posters that coordinate the movement of text and images
- Web pages in a range of styles, such as a minimalist brand-site look, creative visual-focused designs and a tech-product look
- Graphics that stay sharp when enlarged (SVG), and animations of them
- 3D visuals that run in a web browser (Three.js)
- 3D creation using the 3D software Blender, refined through repeated revisions as an agent and output as editable files
SenseTime says the model can also connect to design tools such as Figma, so designers can use AI within the workflows they already use.
Availability and performance
From the day of release, SenseNova 6.8 Flash can be tried in the following places.
- SenseTime's AI assistant Raccoon (小浣熊), where the model can be used for free until December 1, 2026
- SenseNova Studio on the web, where the design features can be tried in a design beta
- SenseNova Token Plan for developers, where it is also available via API
SenseTime says the model outperformed some leading overseas models in areas such as agentic collaboration, data analysis and creating presentation materials. A chart on the official page shows scores on 6 evaluations: on TeamBench, which measures teamwork, and on the data analysis evaluation, it had the highest scores among the models compared. On the remaining 4, including AutomationBench, which measures task automation, Anthropic's Claude Opus and OpenAI's GPT models scored higher. The 2 evaluations for data analysis and slide creation were built in-house by SenseTime.
SenseNova Skills, a collection of skills for office tasks, is also available on GitHub under the MIT license. At the end of October, SenseTime also released SenseNova-Loop-9B-Preview on Hugging Face under the Apache 2.0 license, a model that uses a "looped" design in which the same layers are reused repeatedly. It has about 9 billion parameters; because it reuses the same layers repeatedly, its computation is equivalent to a model of about 14.6 billion parameters.
CoAI's take
When people talk about Chinese AI models, DeepSeek, Qwen and Kimi tend to get the attention, but SenseTime is a company long known for image recognition such as facial recognition, and it is now also focusing on its SenseNova large language models. What stands out about 6.8 Flash is that it puts practical help front and center, carrying documents and designs through to completion, rather than raw performance. Memory across conversations is an area that ChatGPT, Gemini and others are also investing in, which shows that companies are aiming for "AI you can keep handing work to."
On the other hand, the performance figures come from SenseTime's own announcement, and some are from evaluations it built in-house. Its main services are aimed at China, so if you plan to use it from Japan, check the supported regions, pricing and data handling in advance.