SenseTime Records First‑Ever Profit in First Half of 2026 “Models + Token Factory + Agent Harness” Framework Unlocks High‑Value Commercialization
The Group continued to accelerate the large‑scale commercial deployment of its AI capabilities, building system‑level AI capabilities around the framework of “one model system, one Token Factory, and one Agent Harness” to advance AI from intelligence generation toward end‑to‑end task execution and transform these capabilities into scalable commercial value.
Dr. Xu Li, Chairman of the Board and CEO of SenseTime, said: “SenseTime’s three-pillar core capability framework, built around ‘one model system, one Token Factory, and one Agent Harness’, uses a unified model system to continuously push the boundaries of intelligence, leverages infrastructure and optimization capabilities to reduce Token costs and strengthen cost‑based pricing power, and deploys agents deeply integrated into enterprise and individual workflows to deliver complete task execution capabilities. This framework enables SenseTime to move beyond pure price competition and create sustainable commercial value through differentiated technologies and services. At the same time, it allows us to expand our user base, drive recurring revenue growth, and lay a solid foundation for the Group’s high‑quality development.”
One Model System: Advancing Native Multimodal Intelligence and Long‑Horizon Agents with Global Leading Performance
In 2026, SenseTime released a series of models at a rapid pace, spanning native multimodal unification, visual perception and reasoning, controllable generation, long‑horizon agents, spatial intelligence, and physical world models. These included the NEO‑Unify architecture; SenseNova U1, U1.5 and U1.5‑Lite; SenseNova Vision; the office‑focused multimodal agent SenseNova 6.8‑Flash‑Lite; the content creation model SenseNova U1‑Pro; the spatial intelligence model SenseNova SI‑8B; and the Kairos 3.1 world model co‑developed with Ace Robotics. Across multiple core benchmarks, these models matched or outperformed leading international models, including Gemini, GPT, and DeepSeek. Less than four months after the U1 series was open‑sourced, the three projects had collectively garnered more than 11,000 stars on GitHub.
SenseTime is advancing from digital intelligence toward physical AGI through the closed‑loop framework of “architectural innovation, system capabilities, and high‑value delivery,” continuously translating technological breakthroughs into real-world applications and commercial value.
Token Factory: Upgrading from Compute Provisioning to Scaled Intelligence Production
SenseTime’s SenseCore is evolving into a production‑oriented Token Factory. Through integrated model and infrastructure optimization, as well as capabilities in model adaptation, inference optimization, heterogeneous scheduling, energy management and global service delivery, the Token Factory not only expands Token output and lowers unit costs, but also continuously enhances the reasoning, understanding, generation, and execution capabilities embedded in each Token.
In July 2026, SenseCore delivered an average daily Token service volume of 2.4 trillion, representing a YoY increase of approximately 22 times. As of the date of this announcement, the Group’s total deployed computing power had reached 48,000 Petaflops (P).
The Token Factory provides strong support for SenseTime in advancing the performance boundaries of its foundation models and scaling their commercial deployment. It has tripled the training speed for long‑context Agent models, increased image and video generation efficiency sixfold, improved reinforcement learning training efficiency, and shortened the cycle from model R&D validation to large‑scale deployment to just one week. In addition to supporting the SenseNova foundation model, the Token Factory now serves four external foundation model providers, establishing differentiated capabilities across high‑complexity computing domains, including AI for Science (“AI4S”), video generation, embodied intelligence, world models, and city‑level intelligent applications.
In addition, the Group remains committed to advancing the domestic compute ecosystem, increasing the MFU (Model Floating‑Point Utilization) of mainstream domestic chips by up to 2.5 times compared with the baseline levels established by chip manufacturers. SenseNova U1 has achieved Day 0 adaptation across more than ten domestic chip platforms.
Through compute‑energy synergy, the Company continues to reduce the unit cost of intelligence. During the Reporting Period, its Compute‑Energy Synergy Agent achieved load forecasting accuracy of 96% and generated cumulative electricity cost savings of more than RMB12 million. For every 10,000P of computing power, annual carbon emissions were reduced by 24,000 tons.
Agents Accelerate Adoption Across Enterprise and Consumer Markets, Delivering Value Through Workflow Integration and Enhanced Everyday Experiences
Building on its model and Token capabilities, SenseTime connects models, knowledge, tools, and workflows through a unified Agent Harness, transforming underlying AI capabilities into end-to-end task delivery solutions for enterprise and individual users.
The Group’s agents are rapidly expanding across both enterprise and personal workflows. During the Reporting Period, they served more than 1,000 enterprise customers across over 20 key industries, including education, government, marketing, logistics, banking, and insurance. The Office Agent, Raccoon, has served major enterprises including Lenovo, Ping An Technology, China’s three major telecommunications operators, JD.com, and Kylin Software. Total users increased more than fivefold YoY, enterprise users grew more than sixfold, and monthly active users on the cloud platform increased nearly tenfold. In addition, the video‑generation agent Seko reached a daily video production volume of 10,000 minutes. Content created on the platform has amassed an estimated 1.5+ billion views across short‑video platforms, including 10 short-form dramas that have each surpassed 100 million views. AI productivity solutions are also expanding beyond large organizations to serve smaller teams and One‑Person Companies (OPCs). In the consumer market, the Kapi series of agents, covering use cases such as health, finance, and content creation, has surpassed 45 million cumulative users.
Computer Vision remains a strategic gateway for the Group’s expansion into industry verticals and overseas markets. During the Reporting Period, revenue from the business reached RMB500 million, accounting for approximately 17% of the Group’s total revenue. The business extended across more than 20 countries and regions and served more than 4,500 customers worldwide, with recurring customers contributing 67% of revenue.
Bridging Digital and Physical Intelligence, Unlocking Ecosystem Value
SenseTime continued to invest in frontier areas, including world models and embodied intelligence, intelligent terminals, smart healthcare, and chips. The increase in the valuation of certain AI ecosystem investments generated fair‑value gains, contributing to the Group’s first‑ever IFRS profit and reflecting the gradual realization of financial value from our ecosystem assets. Beyond short‑term investment returns, the more significant long‑term value of these ecosystem investments lies in enabling the Group gain early access to frontier domains, continuously acquire real‑world tasks, data, and feedback, and further strengthen SenseTime’s unified multimodal models and system‑level delivery capabilities.
SenseTime is deepening its “Three-pillar Framework”: one model system to raise the ceiling of intelligence and expand agents’ ability to handle complex tasks; one Token Factory to improve quality and efficiency, reduce costs, and enhance global delivery capabilities; and one Agent Harness to integrate multimodal models with the Token Factory, enabling AI to complete end‑to‑end tasks in real-world scenarios.
Looking ahead, SenseTime will continue to advance the iteration of multimodal agents, driving the evolution of the Token Factory from pricing based on Token volume to pricing based on task outcomes. The Group will also deepen the application of multi‑agent collaboration across enterprise, individual, and OPC workflows, delivering efficient and reliable productivity solutions to users.
Hashtag: #SenseTime
The issuer is solely responsible for the content of this announcement.
About SenseTime
Founded in 2014, SenseTime is a leading artificial intelligence (AI) software company, committed to its mission of “creating a better AI-empowered future through original innovation”. The Company continues to lead cutting-edge AI research while accelerating the large-scale commercialization of AI technologies.
Leveraging deep academic expertise and strong in-house innovation capabilities, SenseTime continues to enhance its industry-leading multimodal, multi-task general AI capabilities, while fostering a vibrant developer ecosystem through structured open-source initiatives.
SenseTime’s business spans Generative AI, Vision AI, and emerging innovation-driven segments. Underpinned by SenseCore, its next-generation AI infrastructure, the Company integrates computing power, algorithms, and platforms to enable scalable AI deployment. Building on this foundation, SenseTime has developed the SenseNova foundation model series, delivering cost-efficient capabilities across a wide range of general AI tasks. SenseTime continues to lead the global Vision AI market, ranking No.1 worldwide by market share. Its SenseFoundry platform integrates years of accumulated computer vision expertise with advanced multimodal foundation models, providing stable and efficient Vision AI solutions to industries worldwide.
SenseTime is listed on The Stock Exchange of Hong Kong, with operations spanning more than a dozen countries and regions.
