蚂蚁开源2025外滩大会发布大模型全景图，AI开发现三大趋势：工具、路线与生态分化

Core Insights - The report released by Ant Group and Inclusion AI highlights the rapid evolution of the global AI open-source ecosystem, showcasing China's active role in this field [1][3] Group 1: Open Source Ecosystem - 62% of open-source projects emerged after the "GPT moment" in October 2022, with an average age of only 30 months [3] - Approximately 36,000 global developers contributed to the projects, with the US accounting for 24% and China for 18% of the contributions [3] - Chinese companies prefer open-weight models, while leading US firms tend to adopt closed-source strategies, indicating a clear divergence in open-source approaches [3][6] Group 2: AI Programming Tools - The explosive growth of AI programming tools is a notable trend, significantly enhancing programmer efficiency [4] - New coding tools launched in 2025 average over 30,000 developer stars, with Gemini CLI achieving over 60,000 stars in just three months [4] - The development of these tools is reshaping the software development industry, with a shift towards AI handling repetitive tasks, allowing developers to focus on creative design and complex problem-solving [4][6] Group 3: Key Trends in Model Development - The report identifies several key directions in model development, including a clear division between open-source and closed-source routes in China and the US [6] - Model parameters are trending towards scalability under the MoE architecture, and reinforcement learning is becoming a crucial method for enhancing model inference capabilities [6] - Multi-modal models are accelerating to become mainstream, and model evaluation methods are diversifying into subjective voting and objective assessment [6]