2026 Open Model Landscape Report

According to data from HuggingFace Hub, the scale of the AI ecosystem—including models, datasets, and Spaces—is growing rapidly every day. However, extreme concentration persists, with 85.6% of all models recording fewer than 200 downloads.

• China's Research Institutes Launching Massive Models: Chinese research institutes such as Moonshot, MiniMax, and Xiaomi are bypassing traditional incremental development approaches and launching ultra-large models ranging from 754B to 2.78T parameters, thereby establishing a technological advantage. In contrast, most major US models remain under 130B parameters.

• US Hardware-Centric Ecosystem: Open-source AI participation in the US is led by hardware companies like NVIDIA and AMD. These companies are releasing large volumes of optimized open models to demonstrate the performance of their chips and drive sales, reflecting a trend where the center of open source is shifting from model research labs to hardware and infrastructure companies.

• Mismatch Between Downloads and Interest: A model's 'Likes' count and 'Downloads' count represent different metrics. High download numbers are primarily seen in embedding models or practical small models, illustrating the gap between a model's technical buzz and its actual industrial adoption.