en.Wedoany.com Reported - China's SenseTime officially released its new generation lightweight multimodal agent model—RiriXin SenseNova 6.7 Flash-Lite—on May 8, 2026, simultaneously launching a limited-time free SenseNova Token Plan and open-sourcing the office skillset SenseNova-Skills.
By eliminating the traditional vision-to-text intermediate layer, the new model achieves a leapfrog improvement in agent capabilities with a smaller parameter size, significantly reducing Token consumption during inference. Official data shows that in high-frequency interactive scenarios such as information search, Token consumption drops by 60% compared to text-only agents, with millisecond-level response times. The model adopts a native multimodal architecture, capable of directly understanding complex web layouts, document structures, and financial charts, enabling an integrated "see, think, and do" approach.
To accelerate developer ecosystem building, SenseTime simultaneously launched the limited-time free SenseNova Token Plan. After registration, developers can enjoy a quota of 1,500 barrier-free calls every 5 hours in the first month, experiencing the new model's capabilities at zero cost. The Token Plan fully supports the SenseNova model series, including the newly released SenseNova 6.7 Flash-Lite and the accelerated model SenseNova U1 Fast for infographic generation. Paid packages with tiers such as Lite and Pro will be introduced later, covering different production-level needs from lightweight high-frequency to professional complex tasks.
SenseTime has packaged core capabilities covering high-frequency office scenarios into the SenseNova-Skills skillset, which has been open-sourced on the GitHub platform. The skillset covers typical office capabilities such as image generation, Excel analysis, and deep research, natively supporting mainstream agent frameworks like OpenClaw and Hermes Agent. Developers can either choose the Agent Pack installation package, which integrates the framework and a full suite of plugins for one-click deployment, or obtain the required skill components from GitHub for independent integration. The model features five core agent capabilities: dynamic action decision-making, autonomous toolchain orchestration, noise-resistant environmental perception, autonomous error correction and recovery, and long-sequence memory consistency.
SenseNova 6.7 Flash-Lite is specifically designed for high-frequency production environments, achieving multiple SOTA results at its level in authoritative agent benchmarks, and already supports core workflows across all industries including finance, manufacturing, healthcare, and education. The current competition in the AI field is rapidly evolving from "large parameters" to "lightweight and high efficiency." SenseTime's move to reduce Token consumption with a native multimodal architecture, lower developer barriers with the Token Plan, and build an ecosystem moat with an open-source skillset helps it secure a place in the fierce competition among domestic large models.
This article is compiled by Wedoany. All AI citations must indicate the source as "Wedoany". If there is any infringement or other issues, please notify us promptly, and we will modify or delete it accordingly. Email: news@wedoany.com
