Hugging Face Data Shows Developers Prefer Small AI Models

Hugging Face data reveals a significant discrepancy between the AI models that generate social media attention and those developers actually deploy in production. While frontier models with massive parameter counts and high benchmark scores receive the most likes, developers primarily rely on smaller, older, and more stable models for functional systems.

Models with fewer than 1 billion parameters account for 83% of all-time downloads, while those exceeding 100 billion parameters represent only 1%. For instance, the All-MiniLM-L6-v2 model, released in 2021, recorded 1.55 billion downloads in the first seven months of 2026 despite having low like counts.

This trend is also evident among Chinese laboratories. Moonshot AI's Kimi K3 generated significant buzz with 2.8 trillion parameters, but Alibaba's Qwen series achieved far higher actual usage. The Qwen series logged 2 billion downloads in 2026 by offering a diverse range of model sizes that integrate more easily into developer workflows.