Yating Intelligence’s FedGPT Tops Taiwan’s First Sovereign AI Evaluation
Sovereign artificial intelligence aims to keep models, data and computing infrastructure under local control, reducing exposure of sensitive information while reflecting domestic laws, language and cultural norms. Taiwan’s Ministry of Digital Affairs introduced its first sovereign AI evaluation to assess how well models understand Taiwanese values and perform on local academic subjects, offering government agencies and companies a new benchmark for choosing domestically oriented generative AI systems.
The ministry recently released the results, with FedGPT, developed by Yating Intelligence, ranking first in three categories: Taiwanese values, Chinese-language performance on the General Scholastic Ability Test, and the exam’s social studies section. The 30-billion-parameter, or 30B, model is positioned as a relatively lightweight system designed for fully on-premises deployment. Yating is targeting organizations that require tighter cybersecurity controls, regulatory compliance and responses grounded in Taiwan’s cultural context.
All Coverage
1 original reportsThe Backstory
The history behind this eventYating Intelligence Pitches FedGPT to Tap Taiwan’s On-Premises AI Demand
Taiwan AI Labs, founded in 2017, has promoted trustworthy AI for local industries, while strategic partner Yating Intelligence develops enterprise-grade systems. The push comes as companies focus less on whether generative AI works than on who controls sensitive data, models and computing. In a joint survey, 51% of respondents cited data leakage as a concern, while businesses also flagged unpredictable model updates and dependence on public-cloud providers, making sovereign, on-premises AI a strategic issue.
On June 30, 2026, Yating Intelligence and Taiwan AI Labs released the 2026 Taiwan Enterprise AI Transformation Insights Report, based on 562 questionnaires across more than 10 industries. Only 15% of respondents had deployed AI across multiple departments, while 44% remained in small-scale trials and 41% had yet to adopt it or were still evaluating options. Some 77% saw limits in public-cloud AI and 72% explicitly wanted on-premises deployment. Yating is pitching FedGPT, which keeps data inside an organization and links with existing systems, with meeting-assistant and administrative “universal secretary” tools. No pricing or sales figures were disclosed.
Subscribe to Mark Radar Weekly
Every Friday, the week's strongest signals in your inbox. Unsubscribe anytime.
If you search news on Google, you can set Mark Radar as a preferred source—our coverage will show up more often in your results. Set as preferred source on Google →