Understanding how individuals actually engage with AI models presents a clearer picture than traditional benchmarks, and a recent report by OpenRouter—developed with insights from venture firm Andreessen Horowitz—delivers just that. This comprehensive analysis, which examines more than 100 trillion tokens of anonymized data collected over roughly one year, reveals significant trends in AI usage.
Why is there a surge in open-weight models? The data indicates that these models now comprise approximately one-third of all token volume utilized on the OpenRouter platform. The report identifies a notable increase in the share of Chinese-developed open-weight models. Their representation jumped from around 1.2% of weekly token share in late 2024 to highs close to 30% in various weeks throughout 2025, averaging about 13% over time.
What is meant by open-weight models? While these models provide accessible weights to the public, essential elements such as training data and methodologies usually remain proprietary. This availability allows developers to build upon existing models without disclosing their underlying processes.
What are the primary applications for AI usage? Two prominent use cases have emerged: creative roleplay and coding assistance, with roleplay activities alone accounting for over half of open-weight model interactions. Additionally, a trend involving agentic workflows is on the rise, in which AI models coordinate multiple tasks autonomously rather than responding to single prompts. This development spans both open and proprietary model environments.
Why does this trend hold significance beyond the realm of artificial intelligence? The substantial adoption of open-weight models suggests that competitive barriers for proprietary model providers may not be as robust as once perceived. The increased share of Chinese AI labs tapping into nearly 30% of weekly token interactions through open-weight strategies brings forward discussions surrounding the technology rivalry between the United States and China. Since open-weight models are inherently challenging to regulate via export controls or sanctions, their dominance could shift the landscape of AI development internationally.
The collaboration between OpenRouter and Andreessen Horowitz underscores the strategic implications of this research. As a leading investor in both the crypto and AI sectors, a16z’s engagement with open-weight usage patterns indicates their anticipation of ongoing opportunities at the intersection of these evolving fields.
Investors should closely monitor these trends as they could signal where future advancements and competitive dynamics in the AI marketplace are headed.