Taiwanese electronics giant Foxconn has unveiled its first large language model (LLM), named FoxBrain, marking a significant milestone in the company’s diversification into artificial intelligence (AI). Built on Meta’s open-source Llama 3.1 architecture, FoxBrain features 70 billion parameters and is optimized for traditional Chinese and Taiwanese language styles. The model was developed in just four weeks using 120 Nvidia H100 GPUs, supported by Nvidia’s Quantum-2 InfiniBand networking and the Taipei-1 supercomputer. Designed initially for internal applications, FoxBrain focuses on manufacturing and supply chain optimization, offering capabilities such as data analysis, decision support, document collaboration, mathematics, reasoning, problem-solving, and code generation.

Foxconn emphasized that FoxBrain’s development prioritized efficient training strategies over excessive computational power. Dr. Yung-Hui Li, Director of the Artificial Intelligence Research Center at Hon Hai Research Institute, explained that the model employs adaptive reasoning reflection to enhance logical reasoning and mathematical capabilities. In benchmark tests like TMMLU+, FoxBrain outperformed other models of similar scale, including Taiwan Llama-3-Taiwan-70B, particularly excelling in mathematics and logical reasoning tasks. Although its performance slightly trails China’s DeepSeek AI model, FoxBrain is approaching world-class standards. This achievement highlights Taiwan’s technological talent and its ability to compete internationally in the AI field.

Foxconn plans to expand FoxBrain’s applications beyond internal use by collaborating with technology partners and open-sourcing the model in the future. The company envisions FoxBrain driving advancements in smart manufacturing, electric vehicles (EVs), and smart cities. Nvidia played a crucial role in the development process by providing technical consultation and infrastructure support through its supercomputing resources. Foxconn stated that FoxBrain would serve as an engine to upgrade its three major platforms: Smart Manufacturing, Smart EVs, and Smart Cities. The company intends to showcase FoxBrain’s capabilities at Nvidia’s GTC conference on March 20, where it will present its results under the session titled “From Open Source to Frontier AI: Build, Customize, and Extend Foundation Models.”

The launch of FoxBrain comes amid Foxconn’s strategic shift toward AI and EVs as it faces challenges in its core electronics manufacturing business due to declining smartphone sales. By leveraging AI technologies like FoxBrain, Foxconn aims to enhance operational efficiency and maintain competitiveness in a rapidly evolving industry landscape. The company has expressed confidence in FoxBrain’s ability to transform supply chain management and intelligent decision-making processes. With plans for open-source collaboration and broader implementation across industries, FoxBrain represents a pivotal step in Foxconn’s efforts to integrate cutting-edge AI solutions into its business operations while contributing to Taiwan’s growing prominence in the global AI ecosystem.