uid=SqAUwdd0 RootWebArea "State of AI | OpenRouter" uid=KKyJloLX image "" ---- uid=9MwGdJpV heading "State of AI"

level="1" ✓uid=4Gkkn_El StaticText "State of AI" ✓uid=3BWkaYK0 StaticText "An Empirical 100 Trillion Token Study with OpenRouter" ---- ✓uid=_YT9s2UM StaticText "Malika Aubakirova" ✓uid=TaS56Tls StaticText "Alex Atallah" ✓uid=0iP1sOEf StaticText "Chris Clark" ✓uid=w_cE8Vp9 StaticText "Justin Summerville" ✓uid=Ba8dOHo0 StaticText "Anjney Midha" ---- ✓uid=lQv44JlE StaticText " a16z (Andreessen Horowitz) " ✓uid=6SmOrD5m StaticText " OpenRouter Inc." ---- ✓uid=IfpFJDBn StaticText " Lead contributors. Please see " ✓uid=yHoP-cTc StaticText "Contributions" ✓uid=jnBWCekv StaticText " section for details." ✓uid=KkXUJjXe StaticText "December, 2025" ---- uid=jSfsrvMC heading "Abstract"

level="2" ✓uid=7cM1__7N StaticText "Abstract" paragraph ✓uid=vQ_B-VYd StaticText "The past year has marked a turning point in the evolution and real-world use of large language models (LLMs). With the release of the first widely adopted reasoning model," ---- ✓uid=XiZfD_sE StaticText "o1" ✓uid=wIcUwBl6 StaticText ", on December 5th, 2024, the field shifted from single-pass pattern generation to multi-step deliberation inference, accelerating deployment, experimentation, and new classes of applications. As this shift unfolded at a rapid pace, our empirical understanding of how these models have actually been used in practice has lagged behind. In this work, we leverage the OpenRouter platform, which is an AI inference provider across a wide variety of LLMs, to analyze over 100 trillion tokens of real-world LLM interactions across tasks, geographies, and time. In our empirical study, we observe substantial adoption of open-weight models, the outsized popularity of creative roleplay (beyond just the productivity tasks many assume dominate) and coding assistance categories, plus the rise of agentic inference. Furthermore, our retention analysis identifies " ---- ✓uid=PMMQDINd StaticText "foundational cohorts" ✓uid=NGFqmooP StaticText ": early users whose engagement persists far longer than later cohorts. We term this phenomenon the Cinderella " ---- ✓uid=J_p2fW96 StaticText ""Glass Slipper"" ✓uid=qzPuJLGz StaticText " effect. These findings underscore that the way developers and end-users engage with LLMs "in the wild" is complex and multifaceted. We discuss implications for model builders, AI developers, and infrastructure providers, and outline how a data-driven understanding of usage can inform better design and deployment of LLM systems." ---- uid=1NGSdraI link "Download PDF" ✓uid=4NY6hdAs StaticText "Download PDF" ---- uid=rUO85UpU heading "Introduction"
level="2" ✓uid=57eMk-L0 StaticText "Introduction" uid=X_5r4OEd button ""