Preface

As I sat down this time last year to write the preface to the 2024 China Open Source Annual Report, I wrote: “By 2024, the explosive growth of open source AI has completely reshaped industry norms.” At the time, we looked ahead with great anticipation, yet none of us foresaw that 2025 would turn in a record so far beyond what we had imagined—a year in which open source did more than reshape industry norms; it began to redefine the very rules that govern them.

Early in the year, DeepSeek broke the technological monopoly of the closed-source giants with an open-source model trained through pure reinforcement learning, achieving “model parity.” This breakthrough set off a swift chain reaction: Qwen, GLM, Kimi, and other Chinese open-source models emerged in rapid succession, and for the first time, downloads of Chinese open-source models on Hugging Face overtook those of their American counterparts. Meanwhile, in software, OpenHarmony claimed the top spot in global open-source project influence rankings with an OpenRank score exceeding 60,000, with openEuler close behind. When we speak of “China’s open-source strength” today, we are no longer describing an aspiration. Instead, we are referring to a body of verifiable data.

Yet the most profound question 2025 leaves us with is perhaps not which model has set another record on the leaderboards, but a more fundamental inquiry: when AI starts writing code, what becomes of the underlying logic of open-source collaboration?

The industry has dubbed 2025 “the year of the agent.” GitHub Copilot evolved from a code-completion tool into an autonomous coding agent. Cursor, powered by a product-led growth (PLG) model, surpassed US$1 billion in annual recurring revenue (ARR) within the year. AI coding agents such as Roo-Code, Cline, Aider, and OpenHands proliferated, claiming half the spots on the global ranking of emerging open-source projects. “Vibe coding” went from an internet meme to a genuine way of building software: developers describe their intent in natural language, leaving AI to generate the code, debug it, and even submit the pull request (PR). The barrier to software development has never been lower; productivity has never been higher.

The other side of the coin, however, is already visible in the data. The Data section of this report reveals a warning sign: in 2025, contribution and influence across the global open-source ecosystem, as measured by OpenRank, declined in tandem, with contribution sliding even more sharply. As the report’s analysis notes: “With the widespread adoption of generative AI tools, some tasks that once relied on community collaboration are now being handled by technical means. While this has boosted individual developer productivity, it has, in effect, reduced both the need for interaction among developers and the frequency of their collaboration.” In other words, AI has made individuals stronger, but the collaboration network sparser. When a single developer, with the help of AI, can accomplish alone what once took three to five people working together, issue threads grow shorter and PR reviews lighter. Those seemingly inefficient yet essential processes of knowledge transfer and trust-building within communities are being quietly eroded.

None of this is hyperbole. Open source has never been merely about “making code public”; it has always been about making collaboration happen. If the endgame of AI coding is a world in which every developer becomes a full-stack lone wolf, then the very foundations that sustain open-source communities — their collaborative culture, code review mechanism, mentorship system, and knowledge accumulation — will all face a fundamental restructuring. We do not yet know the answer, but I believe this is precisely the defining question of our time that the open-source communities must confront head-on: the task is not to resist AI, but to think how collaboration should be re-designed for the AI-native era.

This report’s structure has also expanded significantly this year. The Data section introduces, for the first time, the Open Source Development Goals (OSDGs) framework, which brings to light structural differences between the US and Chinese open-source ecosystems. US projects are highly globalized, with domestic contributions accounting for only 38% of the total; Chinese projects, by contrast, remain predominantly domestic, with that share reaching as much as 79%. A truly leading open-source ecosystem must not only export code to the world, but, more importantly, give the world a reason to come in and build alongside it. For the first time, Web 3.0 has a standalone section of its own, examining how Chinese-speaking developers in decentralized open-source protocol communities remain “present in the code, yet absent from the conversation.” Open-Source Embodied AI is likewise new as a standalone section this year. If DeepSeek brought “model parity” to Chinese open source in the digital world, this past year’s embodied-AI wave, led by vision-language-action (VLA) models, world models, and open-source robot platforms, is carrying open source from the digital realm into the physical one and, in doing so, redefining how machines understand the world and act within it. The Open Source Chronicle, organized into nine parts, offers a panoramic view of the year’s milestones, from the State Council’s inclusion of open source in its “AI+” initiative to the second UN Open Source Week. Taken together, they show that open source is evolving from the insider language of technical communities into a shared vocabulary among policymakers and international organizations.

From the vantage point of 2026, 2025 appears to be a watershed. On one side stands the historic rise of China’s open source; on the other, AI’s profound disruption of the very paradigm of open-source collaboration. Prosperity and transformation coexist; certainty and uncertainty intertwine. Yet I remain convinced that what open-source communities do best is turn uncertainty into a starting point for collaboration.

My sincere thanks go to all the volunteers who helped compile this report, and to everyone on this journey who, through their actions, is defining what open source in China means.

— Nadia Jiang, 2025 Chairperson of the Board, KAIYUANSHE