Foxconn, the world’s largest contract electronics manufacturer, has unveiled the first major in-house developed FoxBrain language model with reasoning capability, with the training process taking just four weeks. The new AI model is reportedly based on the Meta Llama 3.1 architecture and optimised for the traditional Chinese dialect used in Taiwan.
Foxconn said that Nvidia supported the project with Taiwan’s largest supercomputer Taipei-1 and technical consulting services during the model training process. During the training of the Foxconn FoxBrain model, 120 Nvidia H100 accelerators were used.
Image source: Foxconn
The range of FoxBrain capabilities developed for the company’s use include data analysis, mathematical computing, problem solving, document collaboration, reasoning and software code generation.
The model ” prioritises optimised learning strategies over simply using computing power” to solve a problem, said Yung-Hui li, director of Foxconn’s AI Research Centre.
Foxconn noted that while FoxBrain is slightly inferior to China’s DeepSeek distillation model, its overall performance is close to global standards, allowing the AI model to be used to improve the company’s production and supply chain management.
In the future, Foxconn plans to open source the model, which will allow it to collaborate with technology partners to expand the model’s application areas, share information and promote AI in manufacturing, supply chain management and intelligent decision-making.
The company said it will reveal additional details about FoxBrain at the Nvidia GTC AI developer conference in mid-March.