On September 13, at the 2025 Inclusion·Bundai Conference AI Open Source Insights Forum, Ant Group Open Source, in collaboration with Inclusion AI, launched the new "Global Large Model Open Source Ecosystem Overview and Trend Report."

This is the 2.0 version of the report, released for the first time in May. It not only comprehensively reveals the current development status and future trends in the field of artificial intelligence open source, but also includes new developments in open source communities over the past hundred days, providing an important reference for industry development.

QQ20250915-103909.png

(Wang Xu, Vice Chairman of the Ant Group Open Source Technology Committee, introduces the global large model open source ecosystem overview and trend report)

Wang Xu, Vice Chairman of the Ant Group Open Source Technology Committee, said: "We objectively present the real situation of the global AI open source ecosystem through data-driven methods. This not only provides references for the industry, but also demonstrates China's important position in the field of AI open source."

Data Reveals New Characteristics of the Large Model Open Source Ecosystem

The report originally came from internal technical trend insights within Ant Group. All the data comes from open source communities, and by analyzing projects across the GitHub platform, the OpenRank algorithm was used to screen and rank the projects. The latest large model open source ecosystem map includes 114 of the most popular open source projects across 22 technical fields, divided into two major technical directions: AI Agent and AI Infra.

QQ20250915-103915.png

(Large Model Open Source Map 2.0)

Along with the map, a detailed insight report titled "Revisiting the Global Large Model Open Source Ecosystem and Trends Based on Community Data" was also released. The report points out that 62% of the open source projects in the large model ecosystem were born after the "GPT Moment" in October 2022, with an average age of only 30 months, reflecting the high-speed iteration characteristics of the AI open source ecosystem.

"Open Sharing" or "Closed Source Dominance": A Divergence in Development Paths

The report shows that among the approximately 360,000 global developers participating in the project development, 24% are American developers, 18% are Chinese developers, followed by India (8%), Germany (6%), and the UK (5%). Together, the US and China account for more than 40% of the core forces. More notably, in terms of large model open source strategies, Chinese companies tend to favor open source models with open weights, while leading U.S. companies mostly adopt closed source models.

Wang Xu pointed out: "These projects can be imagined as digital building blocks, allowing developers to freely combine and build new applications. The enthusiasm of China in sharing these building blocks is making the global ecosystem more vibrant."

The Continuous Growth of AI Coding Tools: Accelerating the 'Efficiency Revolution' of Developers

The most notable trend in the report is the explosive growth of AI coding tools. These tools can automatically generate and modify code, significantly improving programmers' efficiency, and have become one of the hottest areas in the open source community. In terms of form, the tools mainly fall into two categories: "command line tools" (such as Google's Gemini CLI) and "integrated development environment plugins" (such as Cline). The former is favored for its lightweight and flexibility, while the latter focuses on integrating the development process.

Data shows that coding tools emerging in 2025 have, on average, received over 30,000 developer stars. Among them, the Gemini CLI open source project, which has been open for just three months, has already exceeded 60,000 stars, becoming one of the fastest-growing projects. Wang Xu's team observed an interesting phenomenon: "Model manufacturers tend to start with command line tools, while teams focusing on user experience prefer integrated development environments. These two approaches are driving a 'revolution' in programming efficiency."

This tool boom reflects the urgent demand of global developers for "AI assistants." The report points out that as the capabilities of large models improve, in the future, programmers may entrust more repetitive tasks to AI tools, shifting their focus to creative design and complex problem-solving. This trend could reshape the division of labor in the software development industry.

Release of the 2025 Large Model Development Timeline Map

Ant Group Open Source also released the 2025 Large Model Development Timeline Map at the forum. This map outlines the timeline of large model releases by major domestic and international manufacturers since January 2025, including open parameter models and closed-source models. Key information such as model parameters and modalities is marked in the map, offering developers and the community a reference to understand the fierce competition among various manufacturers today.

QQ20250915-103922.png

(Large Model Development Timeline Map)

The report highlights several key directions for large model development: clear differentiation between open source and closed source models between China and the U.S., scaling up model parameters under MoE architecture, enhancing model reasoning capabilities through reinforcement learning, multi-modal models becoming the mainstream, and the development of different evaluation modes based on subjective voting and objective assessment.