Global AI Model Trends 2025
Number of Models Doubles as Small Models Dominate
3 Min Read September 4, 2025
TechInsights projects over 2.5 million AI models worldwide by end of 2025, with small models driving adoption and multimodal architectures on the rise.
TechInsights projects that the number of AI models worldwide will reach 2.5 million in 2025, nearly doubling from last year. This growth highlights the accelerating pace of AI model development and deployment, based on TechInsights’ analysis of models registered on Hugging Face.
As of Q3 2025, Natural Language Processing (NLP) remains the largest category, accounting for 62% of models (404,464). However, NLP’s growth has slowed to 13%, while computer vision models (19% of total, 125,116 models) are surging, driven by applications in autonomous systems.
Other categories—reinforcement learning, audio, and multimodal models—collectively make up about 18%. The rise of multimodal AI models, which integrate text, images, and other data types, signals future shifts in both AI software and hardware infrastructure.
A key trend is the move toward small language models (SLMs). Models with fewer than 1 billion parameters represent 52% of all models, and nearly 70% fall under 3 billion parameters. In contrast, large-scale models over 64B parameters remain rare, with those above 256B representing just 0.04%.
This skew toward efficiency underscores the industry’s pivot to smaller, more specialized, and accessible AI models. TechInsights concludes that the AI industry is entering a new phase of practical expansion, with SLMs becoming the new standard.
Access 2025 AI Report
Read the full 2025 AI Model Trends report in the TechInsights Platform for an in-depth look at over 2.5M AI models, including growth, categories, and emerging trends in NLP, computer vision, and multimodal AI.
