Taiwan’s Leading On-Premises AI Model Arrives! Yating Intelligence FedGPT Sweeps National AI Benchmark with Record-Breaking “Triple Crown”

2026.8.25

MODA AIEC Latest Evaluation Results: GSAT Chinese and Social Studies Both Exceed 84%. Yating Intelligence Unveils a 30B Fully On-Premise Architecture, Delivering the Most Taiwan-Savvy, Secure, and Appropriate Solution, Defining a New Benchmark for Enterprise-Grade “Sovereign AI” in Taiwan.

[Press Release] The “AI Evaluation Center (AIEC),” established by the Administration for Digital Industries of the Ministry of Digital Affairs (MODA) in collaboration with the National Institute of Cyber Security (NICS) and the Industrial Technology Research Institute (ITRI), recently released its latest language model benchmark evaluation results. Yating-FedGPT, a large language model developed by Taiwan’s AI pioneer Yating Intelligence, demonstrated phenomenal capabilities in this authoritative national-level evaluation. It achieved a perfect 100% accuracy score in “Taiwanese Values,” alongside outstanding scores of 86.20% in “GSAT Chinese” and 84.00% in “GSAT Social Studies,” setting a supreme new standard for localized AI model evaluations.

Perfect 100% Score Flawlessly Aligns with Taiwanese Values, Forging the Most Trusted “Taiwan Sovereign AI”

As generative AI integrates deeply into enterprise operations and public sector workflows, whether a model aligns with local social values, legal frameworks, and cultural contexts has become a crucial factor for AI adoption. To establish a fair, objective AI evaluation mechanism that comprehends Taiwanese society, culture, and localized knowledge, the AIEC incorporated three key metrics: “Taiwanese Values,” “GSAT Chinese,” and “GSAT Social Studies.”

Yating-FedGPT scored 100.00% in the “Taiwanese Values” category, indicating its high conformity with mainstream Taiwanese social values and local contexts. Compared to some international general-purpose models that show significant score disparities in this same category, Yating-FedGPT exhibits a clear localization advantage, providing Taiwanese enterprises and government agencies with a more reliable generative AI option tailored to local needs.

GSAT Chinese and Social Studies Both Surpass 84%, Leading Taiwan in Deep Traditional Chinese Comprehension

Beyond its flawless performance in Taiwanese Values, Yating-FedGPT also showcases incredible strength in language comprehension and cross-disciplinary comprehensive reasoning. It delivered remarkable results in the GSAT Chinese (86.20%) and Social Studies (84.00%) evaluations, proving its exceptional acuity regarding Traditional Chinese semantics, historical culture, social systems, and intricate logic.

Fully On-Premise Deployment Secures AI Sovereignty; 30B Lean Architecture Builds the Most Secure Enterprise-Grade Agentic AI

Yating Intelligence highlighted that compared to mainstream international open-source models with massive parameters often reaching hundreds of billions, Yating-FedGPT utilizes a lightweight 30B specification—less than one-tenth of their size—yet outperforms many larger, general-purpose models in evaluations related to localization and Chinese comprehension. This represents a significant breakthrough by the R&D team in model architecture optimization and corpus training technologies.

As a fully on-premise Agentic AI platform designed exclusively for enterprises, Yating-FedGPT emphasizes that “data remains internal and does not go to the cloud.” This fundamentally eliminates concerns regarding the leakage of sensitive data, perfectly complying with the strict cybersecurity and regulatory requirements of highly regulated industries such as finance, healthcare, the public sector, and manufacturing. Through the smaller and more precise 30B model scale, enterprises can easily control AI data sovereignty within a privatized, on-premise environment—with minimal energy consumption and reasonable hardware thresholds—creating a proprietary and supremely secure enterprise brain.

Yating Intelligence: Empowering Seamless Enterprise Transformation with the “Most Taiwan-Savvy and Secure” AI Technology

Cheen Liao, Chairman of Yating Intelligence, stated: “Achieving a ‘100% perfect score in Taiwanese Values’ and high scores in two GSAT subjects in the national-level AIEC evaluation is the highest recognition of Yating Intelligence’s long-term dedication to localized Taiwanese semantics and enterprise AI applications. The Yating-FedGPT platform is designed to address enterprise pain points regarding data privacy, regulatory compliance for sensitive information, and local contextual accuracy. Moving forward, Yating Intelligence will continue to build the ‘most Taiwan-savvy and secure’ fully on-premise AI technologies, assisting Taiwanese enterprises in achieving seamless transformation under the premise of zero risk to data security, and leading the new era of Sovereign AI.”

(Data Source: Ministry of Digital Affairs / AIEC Evaluation Center Language Model Benchmark Evaluation Data)