开源人工智能：合作的价值与未来（英）

开 源 人 工 智 能 ： 合 作 的 价 值 与 未 来研 究 报 告Introduction: AI Enters the Open-Source Era.............................................................................................................................................01The Value of Open-Source AI: Cooperation beyond Competition .............................................................................................................................................03Open-Source AI in China: Practices and Experiences .............................................................................................................................................05Open-Source AI in China: The First Response from the Chinese Community.............................................................................................................................................07An Action Agenda to Further Advance Responsible Open-Source AI.............................................................................................................................................12Chapter 1 Introduction: AI Enters the Open-Source EraArtificial intelligence（AI）is entering an open-source era. Open-source AI projects are surging rapidly①, and the performance of the most advanced open-source AI models is quickly closing the gap with their closed-source counterparts. At the same time, lightweight open-source AI models are dramatically reducing training and deployment costs, empowering every sector while meeting diverse, scenario-specific needs. Because of open source, more people are beginning to use, debug, learn from, and even build their own models. Because of open source, AI is going beyond the technology community to become a tool, teacher and partner for ordi-nary workers. Because of open source, our confidence that technological innovation can drive equitable progress for humanity is strengthened.However, we cannot take for granted that open-source AI—or the positive outcomes it promis-es—will naturally materialize. In recent years, controversies surrounding the value and risks of open-source AI have intensified, to the extent that even the very definition of “open-source AI” has become a subject of significant dispute. Artificial intelligence—particularly large language models (LLMs) and large multimodal models (LMMs)—follows a technological and ecological logic distinct from that of traditional software, which means that many questions demand fresh responses and exploration: What is the true value of open source in advancing AI? How should we address the safety risks of open-source AI? How can we ensure its sustainable develop-ment? The governance reforms of open-source AI that are unfolding in the process of seeking these answers have thus become a “big issue,” touching upon the future trajectory of AI’s tech-nological evolution and whether AI can genuinely contribute to the progress of human society.Represented by China’s Qwen, DeepSeek, MiniMax, Tencent Hunyuan and Baidu Ernie series of open-sour