The past year has seen significant changes in AI capabilities and developer behaviors, driven by advanced reasoning models and the rise of AI-native applications. OpenRouter, a platform routing traffic for over five million developers and more than 300 models, has analyzed over 100 trillion tokens of real-world language model usage. This data reveals that AI is shifting from a simple chat interface to an active participant in workflows, with developers increasingly leveraging multi-step reasoning and iterative processes. OpenAI's December 2024 release of the o1 model marked a turning point, allowing for structured, multi-stage inference that streamlines complex tasks. The data highlights several trends. Open-source models, particularly those focused on reasoning, are gaining traction due to their cost-effectiveness. Creative and coding applications are the largest drivers of token usage, emphasizing the role of AI in software development and content production. Notably, user retention is influenced by breakthrough capabilities; once users experience significant improvements, they tend to stick with their new model. These insights are more apparent in real-world interactions than in traditional benchmarks, showcasing the competitive landscape's nuances. As AI matures, success will favor those who adapt to these emerging workflows. The dataset prompts further investigation into user behaviors and tool usage, revealing questions about roles and retention that previous research lacked. The comprehensive study offers a detailed look at how these trends manifest across various applications and industries, making it a valuable resource for anyone involved in AI development or investment.