Kunlun WanweiThe Group announced on November 27 that it has officially launched a series of models with complex thinking and reasoning capabilities -- "Tiangong Model Version 4.0 "o1 (Skywork o1).
According to the introduction, Skywork o1 is released by Kunlun Wanwei Group with "slow thinking and reasoning"The series of models, the first in the countryAbility to reason logically in Chineseof the o1 model. Unlike existing efforts to reproduce OpenAI o1 models, Skywork o1 not only endogenizes the ability to think, plan, and reflect on model outputs, but also dramatically increases the ability to reason compared to ordinary models on standard evaluation sets, giving the model aImprovement in reasoning skills as a result of thought and reflection.
The three models attached to this release are as follows:
- Skywork o1 Open:An open source model based on Llama 3.1 8B , the model in the same ecological position in the open-source model of the review indicators significantly improved to the SOTA level , and unlocked many lightweight models can not solve the complex mathematical tasks . (Open source address:Click here to go）
- Skywork o1 Lite:The model has complete thinking capabilities with better Chinese language support and faster reasoning and thinking. It excels in math, Chinese logic and reasoning type problems.
- Skywork o1 Preview:This is the full version of the reasoning model, with self-developed online reasoning algorithms, which provides a more varied and "deeper" thinking process, better and higher quality reasoning than Skywork o1 Lite.