Today, China Unicom announced the official launch of the Yuanjing Thinking Chain Large Model, developed by its subsidiary, Unicom Data Intelligence Co., Ltd. This model has been fully open-sourced on platforms like GitHub, Modao, and Shizhi, marking a significant milestone for China Unicom in the realm of artificial intelligence.

The Yuanjing Thinking Chain Large Model is touted as the first open-source general thinking chain large model introduced by a central enterprise in China. It exhibits impressive slow thinking capabilities and general reasoning abilities across diverse disciplines and scenarios. The model is designed to adapt to various tasks of differing complexities while minimizing resource consumption, thus demonstrating remarkable performance.

According to China Unicom, the Yuanjing Thinking Chain Large Model has surpassed the performance of leading general language models, including OpenAI's GPT-4 and Deepseek V3, as well as the open-source Tongyi Qianwen QwQ model, based on evaluations from mainstream rankings. This accomplishment underscores the technological prowess and competitiveness of the Yuanjing Thinking Chain Large Model.

The model's adaptive slow thinking is achieved through two primary strategies: task adaptation and difficulty adaptation. In evaluations involving non-inference tasks, the model tends to generate more concise answers without sacrificing accuracy, thereby enhancing response efficiency. When generating longer thinking chain data, it takes into account the complexity of the questions and the length of the responses. The model utilizes reinforcement learning to align answer length with question difficulty, further improving both accuracy and practicality.

The official demonstration highlights the successful application of the Yuanjing Thinking Chain Model in various areas, including physics, chemistry college entrance examination questions, the 24-point game, and logical reasoning challenges. These applications illustrate the model's extensive potential and significant market opportunities.

The introduction of the Yuanjing Thinking Chain Large Model by China Unicom represents not only a crucial advancement in the company's artificial intelligence initiatives but also contributes to the broader development of AI technology both within China and internationally. Looking ahead, China Unicom plans to enhance its investment and research in artificial intelligence, focusing on fostering innovation and the practical application of AI technologies, thereby contributing valuable insights and strength to economic and social development.

For those interested in accessing the model, it is available on the following platforms:

• GitHub: Yuanjing Model on GitHub
• Modao: Yuanjing Model on Modao
• Shizhi: Yuanjing Model on Shizhi

Key Points

1. China Unicom has launched the Yuanjing Thinking Chain Large Model.
2. The model reportedly outperforms leading AI models like GPT-4.
3. It demonstrates advanced reasoning capabilities and efficiency in various tasks.
4. The model is open-sourced, available on platforms like GitHub and Modao.
5. China Unicom aims to further invest in AI technology development.