Youdao Open-Sources Core Engines of 'Confucius' AI Model 4.0

NetEase Youdao announced the global open-source release of the core engines of its 'Confucius' AI model 4.0, including a multimodal model and a text-to-speech (TTS) model. Developers can download, deploy, and customize them for free.

The multimodal model (27B parameters) is designed for education, achieving top-tier performance in visual math problems. Among models of similar size, it excels at solving complex chart-based math questions, with an 81.4% accuracy on Chinese math problems. It uses a refined thinking chain that cuts output length by 43.2%, lowering inference costs.

The TTS model can clone voices and emotions across languages. Upload a Chinese audio clip, and it can speak 14 languages with the same tone and emotion. It takes just 3 seconds to clone a voice with over 97% accuracy and 85% similarity.