Taiwan’s Foxconn announced on Monday the launch of its first large language model, named “FoxBrain.” The company plans to leverage this technology to enhance manufacturing and supply chain management.
FoxBrain was trained using 120 of Nvidia’s H100 GPUs and completed in just four weeks. Foxconn, the world’s largest contract electronics manufacturer, noted that the model is based on Meta’s Llama 3.1 architecture and is also used in the production of Nvidia’s AI servers.
Foxconn stated that FoxBrain is Taiwan’s first large language model with reasoning capabilities, specifically optimized for traditional Chinese and Taiwanese language styles. While the company acknowledged a slight performance gap compared to China’s DeepSeek distillation model, it emphasized that FoxBrain’s overall performance is nearly on par with world-class standards.
Initially created for internal use, FoxBrain supports various functions, including data analysis, decision support, document collaboration, mathematics, reasoning, problem-solving, and code generation.
Foxconn announced plans to collaborate with technology partners to broaden the applications of its model, share its open-source data, and advance AI in manufacturing, supply chain management, and intelligent decision-making.
Nvidia assisted in the process by providing support through its Taiwan-based supercomputer, “Taipei-1,” and offering technical consulting during the model’s training. Taipei-1, the largest supercomputer in Taiwan, is located in Kaohsiung, a southern city on the island, and is operated by Nvidia. Foxconn will reveal more details about the model at Nvidia’s GTC developer conference in mid-March.
