Ministry of Digital Affairs Unveils First Sovereign AI Evaluation Results! Yating Intelligence’s FedGPT Claims Triple Crown Among Taiwanese Models; 30B Lightweight, Fully On-Premise Architecture Pioneers a New Era for Enterprise Sovereign AI
[Taipei] Guided by the Ministry of Digital Affairs (moda), hosted by the Administration for Digital Industries (ADI), and jointly executed by the AI Evaluation Center (AIEC) and the Industrial Technology Research Institute (ITRI), the “Sovereign AI Evaluation × Trustworthy AI Action” results presentation officially took place today (2nd) at Taipei’s digiBlock C. The event brought together cross-ministerial representatives from the Ministry of Justice, the Ministry of Education, and the Financial Supervisory Commission to jointly declare their commitment to building a sound, trustworthy AI industry environment in Taiwan. In the first sovereign AI evaluation results released, leading domestic provider Yating Intelligence’s flagship model, “Yating FedGPT,” delivered an outstanding performance. It claimed a “triple crown” by achieving a perfect 100% score in “Taiwanese Values,” 86.20% in the “High School General Scholastic Ability Test (GSAT) Chinese Subject,” and 84.00% in the “High School GSAT Social Studies Subject,” setting a new benchmark for localized language models in Taiwan with empirical data.
Breaking the Mold with a Lightweight 30B Specification: Creating a Data-Secure, Cost-Controllable “Enterprise Sovereign AI” Facing the wave of generative AI adoption, highly regulated industries and the public sector generally encounter a core pain point: relying on massive international cloud-based models entails the risk of data leakage and loss of sovereignty, whereas opting for full in-house development comes with prohibitive computing and R&D costs.
Yating Intelligence proposed the “Enterprise Sovereign AI” solution. Instead of blindly stacking parameters to compete in general-purpose capabilities, Yating FedGPT adopted a 30B lightweight specification—less than one-tenth the size of mainstream international open-source models—to create a fully on-premise Agentic AI platform designed specifically for enterprise applications. This architecture ensures that “sensitive data remains within the enterprise and completely off the cloud,” eliminating concerns about data leakage at the source. It precisely meets the stringent information security and regulatory compliance requirements of finance, healthcare, the public sector, and advanced manufacturing industries, realizing the practical value of affordable, cost-controllable computing sovereignty.
Eight Years of Cultivating Local Semantic Understanding: “Taiwanese Consciousness” Deeply Rooted in R&D DNA Addressing the outstanding performance in this evaluation, Yating Intelligence Chairman Cheen Liao noted during the event that the results are a tremendous validation, but Yating Intelligence is “not surprised.” Liao recalled that since launching its speech recognition product “Yating Transcriber” in 2017, Yating Intelligence has journeyed through semantic understanding and federated learning, culminating in its current generative AI technology. Over the past eight years, the company has continuously deepened its understanding of Taiwan’s local social culture, legal systems, and industrial context.
Liao emphasized: “What enterprises truly need is not just a highly conversational cloud-based large model, but a fully controllable productivity tool capable of precisely solving problems. From day one, we chose ‘Yating,’ a name deeply representative of everyday Taiwanese people, and insisted on using the standard Taiwanese translation ‘智慧’ (intelligence) rather than the imported term ‘智能.’ Taiwanese consciousness has long been engraved in Yating Intelligence’s DNA.” The high scores in both the GSAT Chinese (86.20%) and Social Studies (84.00%) evaluations serve as concrete proof of the Yating FedGPT model’s high sensitivity to Taiwan’s historical context, social systems, and precise logical deduction.
National-Level Objective Evaluation Sets the Benchmark: Joining Hands with the Ecosystem to Accelerate the Last Mile of Implementation The national-level standards established by AIEC this time incorporated local knowledge, such as “Taiwanese Values” and the “High School GSAT,” into authoritative metrics, providing a fair and transparent third-party verification standard for AI market adoption. Liao stated that the AIEC evaluation completely shatters the past dilemma enterprises faced when selecting models, clearly proving to the market that Taiwan possesses a secure, trustworthy AI option that truly and profoundly understands local regulations and culture.
To accelerate the promotion of technological applications, Yating Intelligence announced it will further expand cooperation with industry ecosystem partners, packaging the model capabilities recognized by this national-level certification into “out-of-the-box” on-premise AI hardware-software integrated solutions. Yating Intelligence looks forward to moving forward hand in hand with industry, government, academia, and research sectors to realize the long-term vision of “keeping enterprise data within the enterprise and keeping Taiwan’s AI in our own hands,” jointly solidifying national-level technological sovereignty.
(Data source: Ministry of Digital Affairs / AIEC Evaluation Center – Language Model Benchmark Evaluation Data)
