LLM Open Source Ranking: China Leads, No Longer "Just Participating"

On April 17 at the Singularity Intelligent Technology Conference, CSDN and multiple institutions released the “Large Model Technology System Comprehensive Open Source Influence Ranking.”

Seeing the rankings, honestly, I was a bit surprised.

Alibaba and Beijing Academy of Artificial Intelligence took the top two spots in the comprehensive ranking. What does this mean? In the global AI open source ecosystem, Chinese institutions are no longer just “participants,” but starting to “lead.”

This reminds me of a scene two years ago. Back then I was still doing algorithms at a big tech company, and every meeting discussing open source, everyone’s first reaction was “how are foreign open source projects doing.” Stable Diffusion, Llama, Falcon—these names basically defined the AI open source landscape then. Chinese open source projects? Sure they existed, but always felt lacking.

But this ranking completely changed that narrative.

The evaluation system is very detailed, divided into four dimensions: data, models, systems, and evaluation—53 sub-indicators total, analyzing 13,541 public data links. At this granularity, I think it’s much more reliable than simple “star count rankings.”

Specifically, Chinese institutions performed particularly well in “data” and “systems” dimensions. What does this mean?

The “data” dimension refers to open source datasets’ quality, scale, and diversity. Looking at the ranking’s data section, Alibaba’s ModelScope and BAAI’s WuDaoCorpora are both at the forefront. I’ve used both datasets—quality is indeed good, and coverage is broad.

The “systems” dimension refers to AI infrastructure and toolchains. Here we must mention DeepSeek’s open source toolchain—their open sourced training framework and inference engine have excellent reputation in the developer community. I tried their inference engine before; speed is indeed fast.

But what I find more interesting is that this ranking signals: AI open source competition is shifting from “models themselves” to “full-stack ecosystems.”

What is a “full-stack ecosystem”? It means not just models, but data, tools, and evaluation systems. An isolated open source model, without supporting data and tools, is hard to truly deploy. China’s ability to lead this time is precisely because of systematic work in “full-stack.”

My personal feeling is this reflects a change in China’s AI industry: from “single-point breakthroughs” to “systematic construction.”

Two years ago, discussing Chinese AI, we focused on individual models: how big this model’s parameters, how that model performs. But now, we start focusing on the entire technology system: where data comes from, whether toolchains are complete, what evaluation standards are.

This shift, I think, is positive. Because AI deployment is never solved by one model alone. You need data, compute, tools, talent. China’s performance in this open source ranking precisely shows we’re starting to invest in these “infrastructure” areas.

Of course, I must be honest: rankings are momentary achievements; open source ecosystem building is a long-term process. Chinese institutions leading this time doesn’t mean they’ll always lead. Foreign open source communities are rapidly evolving too; competition will only intensify.

But at least, this ranking gave us a signal: Chinese AI open source is no longer just “tagging along.”

This, I think, is worth being proud of.

But then again, for developers, rankings are rankings; actual usability requires personal testing. My current suggestion: if you’re working on AI-related projects, pay more attention to these Chinese open source projects. You might discover some surprises.

After all, the meaning of open source isn’t about who ranks first, but whether it can truly solve problems.