The Taiwanese electronics group Foxconn has developed its own Large Language Model (LLM) with enhanced reasoning capabilities. The model, named FoxBrain, was trained in just four weeks and is initially intended for internal applications in data analysis, mathematics, logical thinking, and code generation.
The development was supported by Nvidia, which provided supercomputer capacities in Taiwan as well as technical consultation. Foxconn utilized 120 H100 graphics processors from Nvidia to efficiently train the model.
According to Foxconn, FoxBrain is based on the architecture of Meta Llama 3.1 and is the first Taiwanese LLM with advanced reasoning capabilities. It has been specifically optimized for traditional Chinese used in Taiwan and parts of the Chinese diaspora.
The company plans to release FoxBrain as an open-source model for cooperation with industry partners. The goal is to enable advancements in manufacturing and supply chain management.
Foxconn stated that the performance of FoxBrain is slightly behind the leading models of China's DeepSeek but is approaching the world’s top. Further technical details will be presented at the Nvidia GTC in March.
In view of increasing competition and shrinking margins in the electronics business, Foxconn is increasingly diversifying into future areas such as artificial intelligence and electric vehicles.
