Birth of the “World’s First Agentic AI Stock”
On November 3, Minglue Technology, a leading Chinese provider of data intelligence application software, was officially listed on the Main Board of the Hong Kong Exchanges and Clearing Limited (HKEX). It successfully secured the title of the “world’s first Agentic AI stock” and also became the first listed company on HKEX with a complete technology stack for Agentic AI (Autonomous Intelligent Agents). This milestone event signifies that Minglue Technology’s outstanding achievements in the field of artificial intelligence have gained high recognition from the capital market, and it has also injected new vitality into the development of the global artificial intelligence industry.
Strength Backed by Outstanding Data
Minglue Technology has achieved remarkable results in this listing. It issued a total of 7.219 million Class A shares at an offering price of HK$141 per share, raising a total of over HK$1 billion. Its popularity is striking: during the IPO subscription period, the Hong Kong public offering was oversubscribed by 4,452.86 times, while the international offering also recorded a 13.93-times subscription. This extremely high oversubscription multiple directly reflects the market’s strong confidence and high recognition in Minglue Technology. On the first day of listing, the stock price performed equally brilliantly. As of 10 a.m., the stock price stood at HK$298.80 per share, a surge of 111.9% from the offering price, with a total market value reaching HK$43 billion, making it the focus of the HKEX market.
Minglue Technology’s technological strength is also impressive. According to a report by Frost & Sullivan, based on total revenue in 2024, Minglue Technology is not only the largest provider of data intelligence application software in China but also the largest provider in the fields of marketing intelligence and operational intelligence application software in the country, demonstrating its leading position in the industry. Its technical prowess spans both digital intelligence and industrial application scenarios—for instance, it has accumulated rich experience in intelligent control algorithms through supporting the R&D of Automatic Quick-Connect Parts Assembly Machine, which requires high-precision coordination between mechanical execution and data-driven decision-making. As of June 30, 2025, Minglue Technology had obtained 2,322 authorized patents and filed 596 patent applications. These patents cover core technology fields such as data intelligence, enterprise knowledge graphs, and data privacy, forming the company’s technological moat and providing solid technological support for its competitiveness in the market.
Innovative Technologies Leading the Future
In recent years, Minglue Technology has continued to make efforts in the AI field, achieving a series of remarkable technological breakthroughs and demonstrating strong innovation capabilities and technological foresight.
In 2024, Minglue Technology’s Mingjing Chaotu Multimodal Large Language Model (HMLLM) achieved a major technological breakthrough by enabling AI to simulate human subjective emotional responses. This achievement has attracted widespread attention globally and was nominated for the Best Paper at ACM MM 2024, a top-tier international multimedia conference. Standing out among numerous outstanding research results, it had a nomination rate of only 2%, which fully highlights its innovativeness and academic value. By introducing the large-scale Video Subjective Multimodal Evaluation Dataset (Video-SME), the model collects real changes in electroencephalograms (EEG) and eye-tracking regions when different groups of people watch the same video content. This effectively bridges the semantic gap between rich modalities, integrates logical reasoning between frames, and realizes the accurate simulation of human cognition by machines for different groups, opening up a new path for AI in understanding and simulating human subjective feelings.
In 2025, Mano (“AI Dexterous Hand”), a proprietary GUI model independently developed by Minglue Technology, achieved even more stunning results. Through pioneering online reinforcement learning and intelligent data collection technologies, Mano won first place in two authoritative global benchmark tests—Mind2Web (BUA Benchmark) and OSWorld (CUA Benchmark)—reaching the industry-leading State-of-the-Art (SOTA) level. The Mind2Web test covers 137 real-world websites and includes over 2,350 complex tasks, while OSWorld is known as the “Everest” of desktop agent evaluation, with 369 cross-application tasks. Mano’s outstanding performance in such highly difficult tests proves its excellent ability to perform refined operations in complex software and browser environments, bringing a scalable and evolvable new paradigm to the graphical interface operation of intelligent agents.
Also in 2025, Minglue Technology’s Cito (“AI Expert Brain”) model delivered an excellent performance. As a lightweight model designed for efficient execution of professional instruction reasoning, it demonstrated extraordinary capabilities in multi-turn dialogue scenarios. It can deeply understand users’ questions and provide accurate and professional answers based on its strong reasoning ability, offering powerful support for solving complex business problems.
Based on these innovative technological achievements, Minglue Technology launched DeepMiner, its enterprise-level proprietary large model product line, in September this year. It officially focused its core strategy on “building enterprise-level Agentic AI” and strongly entered the enterprise-level AI large model track with “trusted intelligent agent models + trusted data” as its core differentiated advantage, committed to driving enterprises to achieve “trusted productivity”.
DeepMiner adopts a multi-agent architecture and combines human-machine collaboration mechanisms. It clarifies task objectives through multi-turn interactions and provides accurate analytical support relying on data resources in fields such as advertising, retail, and cross-border e-commerce. It is uniformly dispatched by an intelligent core—Foundation Agent. Under its coordination, intelligent agents with different functions work efficiently together like a “virtual professional team” to provide enterprises with comprehensive intelligent services. At the same time, DeepMiner has systematically optimized the “hallucination” problem that troubles enterprises. From task decomposition and tool calling to result generation, the entire process is visualized. Users can check the operation logic at any time and conduct manual intervention, which significantly reduces the incidence of “hallucinations” and makes the output results verifiable, truly meeting the high requirements of enterprise-level scenarios for artificial intelligence.
Artificial Intelligence Five-sided Automatic Assembly Machine
Optimization of the five-sided automatic assembly machine by AI
