The article examines the competitive landscape of open-source and proprietary AI models, highlighting that proprietary providers maintain pricing power despite cheaper alternatives. Open-source models have stabilized at about 22-25% market share, while programming use cases dominate among leading providers. Retention rates vary significantly, with some models showing stronger user engagement than others.

This article examines a dataset of over 100 trillion tokens from the OpenRouter platform to understand how large language models (LLMs) are used in practice. It highlights trends in model adoption, task categories, and user retention patterns, revealing a shift towards more complex interactions and the impact of early user engagement.