Kunlun Wanwei Group has achieved a significant milestone by officially launching its proprietary "TianGong Model 4.0" o1 version (Skywork o1). This breakthrough signifies another landmark achievement in China's AI sector. As the first domestically developed o1 model equipped with Chinese logical reasoning capabilities, Skywork o1 is set to usher in a new era of intelligence.

According to Kunlun Wanwei Group, Skywork o1 is not only part of a series of models with "slow thinking and reasoning capabilities" but also integrates inherent abilities such as thinking, planning, and reflection into its output. This groundbreaking advancement significantly enhances Skywork o1's reasoning performance on standard evaluation benchmarks compared to conventional models, effectively achieving the goal of improving reasoning through thought and reflection.

During the launch event, Kunlun Wanwei Group also introduced three models based on Skywork o1 to cater to various industries and application scenarios.

First is Skywork o1 Open, an open-source model based on Llama 3.1 8B. This model has significantly improved evaluation metrics within its ecosystem, achieving state-of-the-art (SOTA) industry-leading performance. Notably, Skywork o1 Open can solve complex mathematical tasks that many lightweight models cannot, providing robust support for research and practical applications. (For the open-source repository, please visit Kunlun Wanwei Group's official website)

Next is Skywork o1 Lite, a model that maintains comprehensive thinking abilities while further optimizing Chinese language support and reasoning speed. Whether handling mathematical, Chinese logical, or inferential tasks, Skywork o1 Lite delivers outstanding performance, offering users a smoother and more efficient experience.

Finally, Skywork o1 Preview is the complete version of the inference model released this time, equipped with Kunlun Wanwei Group's self-developed online inference algorithms. Compared to Skywork o1 Lite, the Preview version offers more diverse and in-depth thinking processes and delivers more refined and high-quality reasoning outcomes. The introduction of this version undoubtedly provides a more reliable option for users requiring high-precision reasoning capabilities.