GUI Model Second Only to Claude: MiningLamp Technology's AI-powered Global Marketing Platform Wins CICAS Grand Prize

APAC - English

• APAC - Traditional Chinese

SUZHOU, China, Jan. 28, 2026 /PRNewswire/ -- On January 25th, the finals of the 3rd China's Innovation Challenge on Artificial Intelligence Application Scene (CICAS) concluded in Suzhou, Jiangsu Province. As one of China's most influential AI competitions, this year's event was jointly organized by the Chinese Association for Artificial Intelligence, the Suzhou Municipal People's Government, and Soochow University, and hosted by the CICAS Organizing Committee and Gusu District People's Government. The competition attracted over 3,250 registered teams, with the national finals bringing together 113 top innovative teams from China and abroad, with over 350 participants competing on the same stage.

MiningLamp Technology (2718.HK), a Hong Kong-listed company hailed as "the world's first Agentic AI stock," in collaboration with Peking University, presented the project "Intelligent Platform for Brand Globalization Creative Generation and Emotional Connection Based on Multimodal Large Models." The project won the Grand Prize in the national finals and was selected as a "2025 National Artificial Intelligence Application Scenario Exemplary Case."

MiningLamp Technology (2718.HK), founded in 2006, is a pioneer in China's enterprise-level large language models and data intelligence sector. The company was listed on the Hong Kong Stock Exchange in November 2025 (stock code: 2718.HK), adhering to the core philosophy of "data-driven trustworthy productivity." It continues to innovate in cutting-edge technologies such as Agentic AI, VLA models, and multimodal large models. Its latest self-developed Mano model has achieved SOTA (state-of-the-art) performance in authoritative international benchmarks such as OSWorld and Mind2Web. In the general large model category, Mano's comprehensive performance ranks second only to Claude-Sonnet-4.5, gaining recognition from both academia and industry.

Wu Minghui, Founder, CEO, and CTO of MiningLamp Technology (2718.HK), introduced that in response to the current global trend of brand "going global," MiningLamp Technology's intelligent platform for brand globalization creative generation and emotional connection based on multimodal large models addresses core challenges faced during brand globalization—cultural differences, content localization, and data-driven decision-making. The platform innovatively integrates four core technological capabilities: "Global Multimodal Content Asset Library + Trustworthy Data Collection Technology + Subjective Emotional Perception Analysis + Video Generation Technology," empowering enterprises to conduct scientific and efficient cross-cultural content creation, emotional assessment, and intelligent decision-making.

1. Global Multimodal Content Asset Library

The platform has built a multimodal content asset library covering major global markets, integrating various media forms such as video, images, and text. This provides a rich material foundation for cross-cultural content creation, helping enterprises quickly generate creative content that aligns with target market cultural characteristics and consumer preferences, significantly reducing content localization costs.

2. Trustworthy Data Collection Technology—"Dexterous Hand" Mano

The platform employs MiningLamp Technology's (2718.HK) self-developed AI dexterous hand model Mano, which can operate in various browser environments, simulating human visual perception to accurately identify any software interface elements, "seeing" and "clicking" like a human. Combined with Mano-Parking web data collection technology, users only need to input a URL and requirement description, and the platform can complete professional-grade data collection. This means enterprises can obtain authentic and effective global multi-source data more quickly, laying a solid foundation for market analysis.

• Ranked first in the specialized model category of the OSWorld authoritative benchmark test
• In the general large model category, comprehensive performance is slightly behind Anthropic's Claude-Sonnet-4.5, ranking second
• Also achieved SOTA (state-of-the-art) results in the Mind2Web benchmark test
• 7B parameter version supports private deployment, meeting security requirements for enterprise-level applications

3. Subjective Emotional Perception Analysis—HMLLM Multimodal Large Model

The platform employs MiningLamp Technology's (2718.HK) self-developed Hypergraph Multimodal Large Language Model (HMLLM), a globally innovative emotional analysis large model that can simulate the subjective feelings of target audiences from different cultural backgrounds, ages, and genders when viewing advertising content. This helps enterprises quickly assess emotional responses from audiences of different cultural backgrounds before content distribution, avoiding cultural misunderstandings and marketing mistakes, significantly improving the precision of cross-cultural marketing.

• Breaking through traditional questionnaire survey methods, constructed the ultra-large-scale subjective personalized datasets Video-SME and SPA-ADV by collecting EEG and eye-tracking data from over 10,000 real subjects
• Achieved multimodal subjective indicator modeling for attention, emotion, and cognition, significantly improving marketing content testing efficiency
• Consistency with real human subjective feelings (R²) exceeds 89%, with expert-verified feasibility exceeding 76%
• The technical paper received Best Paper Nomination at the ACM MM 2024 (CCF-A category) international conference

4. AI-Driven Video Generation Technology

Based on deep analysis results, the platform can intelligently generate and optimize marketing video content, achieving an end-to-end automated process from market insights and emotional assessment to content generation. This compresses the traditional content creation cycle from weeks to hours, significantly improving marketing response speed.

In recent years, MiningLamp Technology (2718.HK) has continuously invested in AI technology research and development, publishing over 20 papers in high-level domestic and international journals and conferences. Representative achievements include:

• ACM MM 2024 (CCF-A category): Subjective Indicator Analysis Model (HMLLM) received Best Paper Nomination
• ACM MM 2025: Multimodal Large Model Eye Movement Prediction Method
• TPAMI (SCI Q1): Few-shot Video Instance Segmentation Method
• IJCV (SCI Q1): Image Generation Method
• AAAI 2026 (CCF-A category): Mano Model Compression Technology accepted as Oral presentation

Relying on MiningLamp Technology's (2718.HK) technological accumulation in the AI field, the platform has currently gained recognition from multiple enterprise clients.

At the conference closing ceremony, MiningLamp Technology (2718.HK), as a key cooperation project partner, reached a cooperation intention with Gusu District to collaborate on AI technology research and development and scenario application implementation.

MiningLamp Technology (2718.HK) stated that the company would continue to adhere to the core philosophy of "data-driven trustworthy productivity," starting with the "brand globalization" scenario and relying on globally leading Agentic AI technology to accelerate the application of trustworthy AI in more vertical industries. As the brand globalization process accelerates, MiningLamp Technology's (2718.HK) brand globalization intelligent platform is expected to become an important technological support for enterprises to enhance their global competitiveness.

SOURCE Mininglamp Technology