SenseTime (商汤科技) launches SenseNova 6.7 Flash-Lite, a lightweight multimodal agent model, alongside...
SenseTime (商汤科技) launches SenseNova 6.7 Flash-Lite, a lightweight multimodal agent model, alongside a free token plan and open-sourced office skill sets. The model claims a 60% reduction in token consumption for information search tasks by eliminating the visual-to-text intermediate layer, achieving native multimodal reasoning on web layouts, documents, and financial charts. The release includes a token plan offering 1,500 free calls every five hours for the first month, and the SenseNova-Skills suite has been open-sourced on GitHub.
This launch matters for the AI agent segment because it exemplifies the emerging pattern of native multimodal agent models designed to minimize inference cost and latency for real-world workflows. SenseTime is directly targeting cost-sensitive enterprise use cases such as data analysis, automated reporting, and presentation generation, positioning the Flash-Lite variant as a workhorse for high-frequency production environments. The 60% token reduction claim, if validated, signals a structural shift in how multimodal agents handle inference economics — a critical lever for enterprise adoption.
The move also updates our understanding of SenseTime's position in the Chinese foundation-model race: unlike peers focused on scale, SenseTime is doubling down on efficiency and agentic task completion. The open-source release of SenseNova-Skills and the generous free-tier token plan mirror the hyperscaler-distribution pattern seen in other regions, aiming to build developer mindshare before monetizing at scale. However, the fierce competition from ByteDance's Doubao, Baidu's ERNIE Bot, and Zhipu AI means SenseTime must deliver consistent quality to convert trial users into paying customers.
