In recent open-source AI large model evaluations, China's open-source AI technology once again demonstrated its formidable strength. The models DeepSeek, Qwen, and Kimi were rated as joint first in influence, a news that has attracted widespread attention in the industry. This ranking, released by AI researchers Nathan Lambert and Florian Brand, covers 35 institutions, more than half of which are Chinese teams. This highlights the rapid development of China in the open-source field, standing in sharp contrast to the closed-source choices of American companies.
In the evaluation, the three models hailed as "pioneers" showed remarkable performance. DeepSeek's R1 performed excellently in multiple tests, even surpassing some top closed-source models, establishing its position in open-source AI. Alibaba's Qwen series has successfully generated dozens of cross-domain models, covering the needs of multiple industries. At the same time, Kimi also launched the world's first open-source large model with trillions of parameters, further solidifying its influence in the industry.
Notably, the second-tier models Zhipu and MiniMax also come from within the country, showcasing China's comprehensive layout in open-source AI model development. Meanwhile, American open-source models performed modestly, with OpenAI's models only ranking in the fourth tier. Last year's popular Meta Llama3 dropped to the bottom of the list, surprising many. According to recent news, Meta may also abandon its subsequent open-source plans, prompting reflections on the changes in the industry.
Currently, Chinese companies are gradually leading the trend in the open-source AI field, thanks to their strong technological accumulation and market adaptability. Meanwhile, American open-source models not only have fewer numbers but also significantly less influence. How this situation will evolve remains something we should continue to pay attention to.
