SenseTime Launches SenseNova 6.7 Flash-Lite, Achieving a Leapforward Upgrade in Multimodal Capabilities

The competition in the field of artificial intelligence is rapidly evolving from "large parameters" to "lightweight and high efficiency." SenseTime has officially launched its new lightweight multimodal intelligent agent model—SenseNova 6.7 Flash-Lite today. Designed specifically for "real-world workflows," this model breaks through the bottlenecks of traditional agents in handling complex long-chain tasks through innovative native multimodal architecture.

The core advantage of this model lies in its excellent visual understanding and logical decision-making capabilities. Unlike previous methods that relied on an intermediate "visual-to-text" layer, SenseNova 6.7 Flash-Lite can directly read complex web layouts, document structures, and financial charts. This integrated "see, think, act" mechanism enables the model to achieve a high success rate in high-difficulty office scenarios such as data analysis, in-depth research, and automated PPT generation.

In practical production applications, efficiency and cost are key concerns for enterprises. Official data shows that by eliminating the intermediate conversion process, the model maintains a small parameter size while achieving leading-level intelligent agent capabilities. In high-frequency interaction scenarios such as information search, its Token consumption is reduced by approximately 60% compared to pure text-based intelligent agents, and it can achieve millisecond-level response feedback.

To further lower the entry barrier for developers and promote ecosystem growth, SenseTime has also launched the SenseNova Token Plan limited-time free activity. Developers who participate in the activity can get an unrestricted usage quota of 1500 times, refreshed every 5 hours, in their first month. In addition, SenseTime has encapsulated the core office skills of its model into SenseNova-Skills and officially open-sourced it on the GitHub platform. This move aims to help developers more conveniently build high-performance intelligent agent applications through technology sharing.