!function(){try{var e="undefined"!=typeof window?window:"undefined"!=typeof global?global:"undefined"!=typeof globalThis?globalThis:"undefined"!=typeof self?self:{},n=(new e.Error).stack;n&&(e._posthogChunkIds=e._posthogChunkIds||{},e._posthogChunkIds[n]="01a0879f-7ac3-7280-89a6-dfff314ac3f8")}catch(e){}}();(globalThis.TURBOPACK||(globalThis.TURBOPACK=[])).push(["object"==typeof document?document.currentScript:void 0,457357,e=>{"use strict";var t=e.i(222666),n=e.i(505962),o=e.i(553215),s=e.i(712621),r=e.i(252358),i=e.i(596694),a=e.i(831186);function l(e){let o,s=(0,n.c)(4),{id:r,children:i,className:a}=e,l=`mt-8 ${a??""}`;return s[0]!==i||s[1]!==r||s[2]!==l?(o=(0,t.jsx)("div",{id:r,className:l,children:i}),s[0]=i,s[1]=r,s[2]=l,s[3]=o):o=s[3],o}e.s(["DataAndMethodology",0,function(){let e,c,d,m,h,u,f,p,g,x,y,b,v,j,_,w,S,N,k,C,A,L,T,I,R,P,M,E,O,D,H,z,W,U,F,B,G,q,X,Y,$,J,K,V,Z,Q,ee,et,en,eo,es,er,ei,ea,el,ec,ed,em,eh,eu,ef,ep,eg,ex,ey,eb,ev,ej=(0,n.c)(71),e_=(0,r.usePosthogClient)();return ej[0]===Symbol.for("react.memo_cache_sentinel")?(e=(0,t.jsx)(i.Heading,{level:2,children:"Data and Methodology"}),ej[0]=e):e=ej[0],ej[1]===Symbol.for("react.memo_cache_sentinel")?(c=(0,t.jsx)(i.Heading,{level:3,children:"OpenRouter Platform and Dataset"}),ej[1]=c):c=ej[1],ej[2]===Symbol.for("react.memo_cache_sentinel")?(d=(0,t.jsx)("strong",{children:"OpenRouter"}),ej[2]=d):d=ej[2],ej[3]===Symbol.for("react.memo_cache_sentinel")?(m=(0,t.jsxs)("p",{children:["Our analysis is based on metadata collected from the ",d," ",'platform, a unified AI inference layer that connects users and developers to hundreds of large language models. Each user request on OpenRouter is executed against a user-selected model, and structured metadata describing the resulting "generation" event is logged. The dataset used in this study consists of'," ",(0,t.jsx)("strong",{children:"anonymized request-level metadata"})," for billions of prompt–completion pairs from a global user base, spanning approximately two years up to the time of writing. We do zoom in on the last year."]}),ej[3]=m):m=ej[3],ej[4]===Symbol.for("react.memo_cache_sentinel")?(h=(0,t.jsx)("strong",{children:(0,t.jsx)("em",{children:"metadata"})}),ej[4]=h):h=ej[4],ej[5]===Symbol.for("react.memo_cache_sentinel")?(u=(0,t.jsxs)("p",{children:["Crucially, we did not have access to the underlying text of prompts or completions. Our analysis relies entirely on"," ",h," ","that capture the structure, timing, and context of each ",(0,t.jsx)("em",{children:"generation"}),", without exposing user content. This privacy-preserving design enables large-scale behavioral analysis."]}),f=(0,t.jsx)("p",{children:"Each generation record includes information on timing, model and provider identifiers, token usage, and system performance metrics. Token counts encompass both prompt (input) and completion (output) tokens, allowing us to measure overall model workload and cost. Metadata also include fields related to geographic routing, latency, and usage context (for example, whether the request was streamed or cancelled, or whether tool-calling features were invoked). Together, these attributes provide a detailed but non-textual view of how models are used in practice."}),ej[5]=u,ej[6]=f):(u=ej[5],f=ej[6]),ej[7]===Symbol.for("react.memo_cache_sentinel")?(p=(0,t.jsxs)("p",{children:["All analyses, aggregations, and most visualizations based on this metadata were conducted using the ",(0,t.jsx)("strong",{children:"Hex"})," analytics platform, which provided a reproducible pipeline for versioned SQL queries, transformations, and final figure generation."]}),ej[7]=p):p=ej[7],ej[8]===Symbol.for("react.memo_cache_sentinel")?(g=(0,t.jsxs)(l,{id:"openrouter-platform",children:[c,(0,t.jsxs)("div",{className:"prose max-w-none",children:[m,u,f,p,(0,t.jsxs)("p",{children:["We emphasize that this dataset is ",(0,t.jsx)("strong",{children:"observational"}),": it reflects real-world activity on the OpenRouter platform, which itself is shaped by model availability, pricing, and user preferences. As of 2025, OpenRouter supports more than 300+ active models from over 70 providers and serves millions of developers and end-users, with over 50% of usage originating outside the United States. While certain usage patterns outside the platform are not captured, OpenRouter's global scale and diversity make it a representative lens on large-scale LLM usage dynamics."]})]})]}),ej[8]=g):g=ej[8],ej[9]===Symbol.for("react.memo_cache_sentinel")?(x=(0,t.jsx)(i.Heading,{level:3,children:"GoogleTagClassifier for Content Categorization"}),ej[9]=x):x=ej[9],ej[10]===Symbol.for("react.memo_cache_sentinel")?(y=(0,t.jsx)("strong",{children:"OpenRouter performs internal categorization on a random sample comprising approximately 0.25% of all prompts"}),ej[10]=y):y=ej[10],ej[11]===Symbol.for("react.memo_cache_sentinel")?(b=(0,t.jsx)("strong",{children:"GoogleTagClassifier"}),ej[11]=b):b=ej[11],ej[12]!==e_?(v=()=>{e_.capture(o.PostHogEvent.ClickStateOfAIExternalLink,{url:"https://cloud.google.com/natural-language/docs/classifying-text",link_text:"classifyText content-classification API",section:"data-and-methodology"})},ej[12]=e_,ej[13]=v):v=ej[13],ej[14]===Symbol.for("react.memo_cache_sentinel")?(j=(0,t.jsx)("code",{children:"classifyText"}),ej[14]=j):j=ej[14],ej[15]!==v?(_=(0,t.jsxs)("p",{children:["No direct access to user prompts or model outputs was available for this study. Instead,"," ",y," ","and responses through a non-proprietary module ",b,". While this represents only a fraction of total activity, the underlying dataset remains substantial given the overall query volume processed by OpenRouter. GoogleTagClassifier interfaces with Google Cloud Natural Language's"," ",(0,t.jsxs)(s.ExternalLink,{href:"https://cloud.google.com/natural-language/docs/classifying-text",target:"_blank",rel:"noreferrer",className:"text-sm focus-visible:border-focus-border focus-visible:shadow-focus focus-visible:outline-none",onClick:v,children:[j," content-classification API"]})]}),ej[15]=v,ej[16]=_):_=ej[16],ej[17]===Symbol.for("react.memo_cache_sentinel")?(w=(0,t.jsx)("code",{children:"/Computers & Electronics/Programming"}),ej[17]=w):w=ej[17],ej[18]===Symbol.for("react.memo_cache_sentinel")?(S=(0,t.jsxs)("p",{children:["The API applies a hierarchical, language-agnostic taxonomy to textual input, returning one or more category paths (e.g., ",w,","," ",(0,t.jsx)("code",{children:"/Arts & Entertainment/Roleplaying Games"}),") with corresponding confidence scores in the range [0,1]. The classifier operates directly on prompt data (up to the first 1,000 characters). The classifier is deployed within OpenRouter's infrastructure, ensuring that classifications remain anonymous and are not linked to individual customers. Categories with confidence scores below the default threshold of 0.5 are excluded from further analysis. The classification system itself operates entirely within OpenRouter's infrastructure and was not part of this study; our analysis relied solely on the resulting categorical outputs (effectively metadata describing prompt classifications) rather than the underlying prompt content."]}),ej[18]=S):S=ej[18],ej[19]===Symbol.for("react.memo_cache_sentinel")?(N=(0,t.jsx)("em",{children:"tags"}),ej[19]=N):N=ej[19],ej[20]===Symbol.for("react.memo_cache_sentinel")?(k=(0,t.jsxs)("p",{children:["To make these fine-grained labels useful at scale, we map GoogleTagClassifier's taxonomy to a compact set of study-defined buckets and assign each request ",N,". Each tag rolls up to higher level ",(0,t.jsx)("em",{children:"category"})," in one to one way. Representative mappings include:"]}),ej[20]=k):k=ej[20],ej[21]===Symbol.for("react.memo_cache_sentinel")?(C=(0,t.jsx)("strong",{children:"Programming:"}),ej[21]=C):C=ej[21],ej[22]===Symbol.for("react.memo_cache_sentinel")?(A=(0,t.jsx)("code",{className:"rounded-sm bg-card-hover px-1.5 py-0.5 font-mono text-sm",children:"/Computers & Electronics/Programming"}),ej[22]=A):A=ej[22],ej[23]===Symbol.for("react.memo_cache_sentinel")?(L=(0,t.jsxs)("li",{className:"mt-2 first:mt-0",children:[C," from"," ",A," ","or"," ",(0,t.jsx)("code",{className:"rounded-sm bg-card-hover px-1.5 py-0.5 font-mono text-sm",children:"/Science/Computer Science/*"})]}),ej[23]=L):L=ej[23],ej[24]===Symbol.for("react.memo_cache_sentinel")?(T=(0,t.jsx)("strong",{children:"Roleplay:"}),ej[24]=T):T=ej[24],ej[25]===Symbol.for("react.memo_cache_sentinel")?(I=(0,t.jsx)("code",{className:"rounded-sm bg-card-hover px-1.5 py-0.5 font-mono text-sm",children:"/Games/Roleplaying Games"}),ej[25]=I):I=ej[25],ej[26]===Symbol.for("react.memo_cache_sentinel")?(R=(0,t.jsxs)("li",{className:"mt-2 first:mt-0",children:[T," from"," ",I," ","and creative dialogue leaves under"," ",(0,t.jsx)("code",{className:"rounded-sm bg-card-hover px-1.5 py-0.5 font-mono text-sm",children:"/Arts & Entertainment/*"})]}),ej[26]=R):R=ej[26],ej[27]===Symbol.for("react.memo_cache_sentinel")?(P=(0,t.jsx)("strong",{children:"Translation:"}),ej[27]=P):P=ej[27],ej[28]===Symbol.for("react.memo_cache_sentinel")?(M=(0,t.jsxs)("li",{className:"mt-2 first:mt-0",children:[P," from"," ",(0,t.jsx)("code",{className:"rounded-sm bg-card-hover px-1.5 py-0.5 font-mono text-sm",children:"/Reference/Language Resources/*"})]}),ej[28]=M):M=ej[28],ej[29]===Symbol.for("react.memo_cache_sentinel")?(E=(0,t.jsx)("strong",{children:"General Q&A / Knowledge:"}),ej[29]=E):E=ej[29],ej[30]===Symbol.for("react.memo_cache_sentinel")?(O=(0,t.jsx)("code",{className:"rounded-sm bg-card-hover px-1.5 py-0.5 font-mono text-sm",children:"/Reference/General Reference/*"}),ej[30]=O):O=ej[30],ej[31]===Symbol.for("react.memo_cache_sentinel")?(D=(0,t.jsxs)("li",{className:"mt-2 first:mt-0",children:[E," from"," ",O," ","and"," ",(0,t.jsx)("code",{className:"rounded-sm bg-card-hover px-1.5 py-0.5 font-mono text-sm",children:"/News/*"})," ","when the intent appears to be factual lookup"]}),ej[31]=D):D=ej[31],ej[32]===Symbol.for("react.memo_cache_sentinel")?(H=(0,t.jsx)("strong",{children:"Productivity/Writing:"}),ej[32]=H):H=ej[32],ej[33]===Symbol.for("react.memo_cache_sentinel")?(z=(0,t.jsx)("code",{className:"rounded-sm bg-card-hover px-1.5 py-0.5 font-mono text-sm",children:"/Computers & Electronics/Software/Business & Productivity Software"}),ej[33]=z):z=ej[33],ej[34]===Symbol.for("react.memo_cache_sentinel")?(W=(0,t.jsxs)("li",{className:"mt-2 first:mt-0",children:[H," from"," ",z," ","or"," ",(0,t.jsx)("code",{className:"rounded-sm bg-card-hover px-1.5 py-0.5 font-mono text-sm",children:"/Business & Industrial/Business Services/Writing & Editing Services"})]}),ej[34]=W):W=ej[34],ej[35]===Symbol.for("react.memo_cache_sentinel")?(U=(0,t.jsx)("strong",{children:"Education:"}),ej[35]=U):U=ej[35],ej[36]===Symbol.for("react.memo_cache_sentinel")?(F=(0,t.jsxs)("li",{className:"mt-2 first:mt-0",children:[U," from"," ",(0,t.jsx)("code",{className:"rounded-sm bg-card-hover px-1.5 py-0.5 font-mono text-sm",children:"/Jobs & Education/Education/*"})]}),ej[36]=F):F=ej[36],ej[37]===Symbol.for("react.memo_cache_sentinel")?(B=(0,t.jsx)("strong",{children:"Literature/Creative Writing:"}),ej[37]=B):B=ej[37],ej[38]===Symbol.for("react.memo_cache_sentinel")?(G=(0,t.jsx)("code",{className:"rounded-sm bg-card-hover px-1.5 py-0.5 font-mono text-sm",children:"/Books & Literature/*"}),ej[38]=G):G=ej[38],ej[39]===Symbol.for("react.memo_cache_sentinel")?(q=(0,t.jsxs)("li",{className:"mt-2 first:mt-0",children:[B," from"," ",G," ","and narrative leaves under"," ",(0,t.jsx)("code",{className:"rounded-sm bg-card-hover px-1.5 py-0.5 font-mono text-sm",children:"/Arts & Entertainment/*"})]}),ej[39]=q):q=ej[39],ej[40]===Symbol.for("react.memo_cache_sentinel")?(X=(0,t.jsx)("strong",{children:"Adult:"}),ej[40]=X):X=ej[40],ej[41]===Symbol.for("react.memo_cache_sentinel")?(Y=(0,t.jsxs)("li",{className:"mt-2 first:mt-0",children:[X," from"," ",(0,t.jsx)("code",{className:"rounded-sm bg-card-hover px-1.5 py-0.5 font-mono text-sm",children:"/Adult"})]}),ej[41]=Y):Y=ej[41],ej[42]===Symbol.for("react.memo_cache_sentinel")?($=(0,t.jsxs)("ul",{className:"[&_code]:before:content-none [&_code]:after:content-none",children:[L,R,M,D,W,F,q,Y,(0,t.jsxs)("li",{className:"mt-2 first:mt-0",children:[(0,t.jsx)("strong",{children:"Others:"})," for the long tail of prompts when no dominant mapping applies. (Note: we omit this category from most analyses below.)"]})]}),ej[42]=$):$=ej[42],ej[43]===Symbol.for("react.memo_cache_sentinel")?(J=(0,t.jsx)("em",{children:"how much"}),ej[43]=J):J=ej[43],ej[44]===Symbol.for("react.memo_cache_sentinel")?(K=(0,t.jsxs)("p",{children:["There are inherent limitations to this approach, for instance, reliance on a predefined taxonomy constrains how novel or cross-domain behaviors are categorized, and certain interaction types may not yet fit neatly within existing classes. In practice, some prompts receive multiple category labels when their content spans overlapping domains. Nonetheless, the classifier-driven categorization provides us with a lens for downstream analyses. This enables us to quantify not just ",J," LLMs are used but"," ",(0,t.jsx)("em",{children:"what for"}),"."]}),ej[44]=K):K=ej[44],ej[45]!==_?(V=(0,t.jsxs)(l,{id:"google-tagclassifier",children:[x,(0,t.jsxs)("div",{className:"prose max-w-none",children:[_,S,k,$,K]})]}),ej[45]=_,ej[46]=V):V=ej[46],ej[47]===Symbol.for("react.memo_cache_sentinel")?(Z=(0,t.jsx)(i.Heading,{level:3,children:"Model and Token Variants"}),ej[47]=Z):Z=ej[47],ej[48]===Symbol.for("react.memo_cache_sentinel")?(Q=(0,t.jsx)("p",{children:"A few variants are worth explicitly calling out:"}),ee=(0,t.jsx)("em",{children:"Open Source vs. Proprietary:"}),ej[48]=Q,ej[49]=ee):(Q=ej[48],ee=ej[49]),ej[50]===Symbol.for("react.memo_cache_sentinel")?(et=(0,t.jsx)("strong",{children:"open source (OSS, for simplicity)"}),ej[50]=et):et=ej[50],ej[51]===Symbol.for("react.memo_cache_sentinel")?(en=(0,t.jsxs)("li",{children:[ee," We label models as"," ",et," if their weights are publicly available, and ",(0,t.jsx)("strong",{children:"closed source"})," if access is only via a restricted API (e.g., Anthropic's Claude). This distinction lets us measure adoption of community-driven models versus proprietary ones."]}),eo=(0,t.jsx)("em",{children:"Origin (Chinese vs. Rest-of-World):"}),ej[51]=en,ej[52]=eo):(en=ej[51],eo=ej[52]),ej[53]===Symbol.for("react.memo_cache_sentinel")?(es=(0,t.jsx)("strong",{children:"Chinese models"}),ej[53]=es):es=ej[53],ej[54]===Symbol.for("react.memo_cache_sentinel")?(er=(0,t.jsxs)("li",{children:[eo," Given the rise of Chinese LLMs and their distinct ecosystems, we tag models by primary locale of development."," ",es," include those developed by organizations in China, Taiwan, or Hong Kong (e.g., Alibaba's Qwen, Moonshot AI's Kimi, or DeepSeek)."," ",(0,t.jsx)("strong",{children:"RoW (Rest-of-World) models"})," cover North America, Europe, and other regions."]}),ei=(0,t.jsx)("em",{children:"Prompt vs. Completion Tokens:"}),ej[54]=er,ej[55]=ei):(er=ej[54],ei=ej[55]),ej[56]===Symbol.for("react.memo_cache_sentinel")?(ea=(0,t.jsx)("strong",{children:"prompt tokens"}),ej[56]=ea):ea=ej[56],ej[57]===Symbol.for("react.memo_cache_sentinel")?(el=(0,t.jsx)("strong",{children:"completion tokens"}),ej[57]=el):el=ej[57],ej[58]===Symbol.for("react.memo_cache_sentinel")?(ec=(0,t.jsx)("strong",{children:"Total tokens"}),ej[58]=ec):ec=ej[58],ej[59]===Symbol.for("react.memo_cache_sentinel")?(ed=(0,t.jsx)("strong",{children:"Reasoning tokens"}),ej[59]=ed):ed=ej[59],ej[60]===Symbol.for("react.memo_cache_sentinel")?(em=(0,t.jsxs)("ul",{children:[en,er,(0,t.jsxs)("li",{children:[ei," We distinguish between"," ",ea,", which represent the input text provided to a model, and ",el,", which represent the model's generated output."," ",ec," equal the sum of prompt and completion tokens."," ",ed," represent internal reasoning steps in models with native reasoning capabilities and are included within"," ",(0,t.jsx)("strong",{children:"completion tokens"}),"."]})]}),ej[60]=em):em=ej[60],ej[61]===Symbol.for("react.memo_cache_sentinel")?(eh=(0,t.jsx)("strong",{children:"token volume"}),ej[61]=eh):eh=ej[61],ej[62]===Symbol.for("react.memo_cache_sentinel")?(eu=(0,t.jsxs)(l,{id:"model-variants",children:[Z,(0,t.jsxs)("div",{className:"prose max-w-none",children:[Q,em,(0,t.jsxs)("p",{children:["Unless otherwise noted, ",eh," refers to"," ",(0,t.jsx)("strong",{children:"the sum of prompt (input) and completion (output) tokens"}),"."]})]})]}),ej[62]=eu):eu=ej[62],ej[63]===Symbol.for("react.memo_cache_sentinel")?(ef=(0,t.jsx)(i.Heading,{level:3,children:"Geographic Segmentation"}),ej[63]=ef):ef=ej[63],ej[64]===Symbol.for("react.memo_cache_sentinel")?(ep=(0,t.jsxs)(l,{id:"geographic-segmentation",children:[ef,(0,t.jsxs)("div",{className:"prose max-w-none",children:[(0,t.jsxs)("p",{children:["To understand regional patterns in LLM usage, we segment requests by user geography. Direct request metadata (like IP-based location) is typically imprecise or anonymized. Instead, we determine user region based on the ",(0,t.jsx)("strong",{children:"billing location"})," ","associated with each account. This provides a more reliable proxy for user geography, as billing data reflects the country or region linked to the user's payment method or account registration. We use this billing-based segmentation in our analysis of regional adoption and model preferences."]}),(0,t.jsx)("p",{children:"This method has limitations. Some users employ third-party billing or shared organizational accounts, which may not correspond to their actual location. Enterprise accounts may aggregate activity across multiple regions under one billing entity. Despite these imperfections, billing geography remains the most stable and interpretable indicator available for privacy-preserving geographic analysis given the metadata we had access to."})]})]}),ej[64]=ep):ep=ej[64],ej[65]===Symbol.for("react.memo_cache_sentinel")?(eg=(0,t.jsx)(i.Heading,{level:3,children:"Time Frame and Coverage"}),ej[65]=eg):eg=ej[65],ej[66]===Symbol.for("react.memo_cache_sentinel")?(ex=(0,t.jsx)("em",{children:"Programming"}),ej[66]=ex):ex=ej[66],ej[67]===Symbol.for("react.memo_cache_sentinel")?(ey=(0,t.jsx)("em",{children:"Roleplay"}),ej[67]=ey):ey=ej[67],ej[68]===Symbol.for("react.memo_cache_sentinel")?(eb=(0,t.jsxs)(l,{id:"time-frame",children:[eg,(0,t.jsxs)("div",{className:"prose max-w-none",children:[(0,t.jsxs)("p",{children:["Our analyses primarily cover a rolling 13-month period ending on November, 2025, but not all underlying metadata spans this full window. Most model-level and pricing analyses were focused on November 3, 2024 – November 30, 2025 time frame. However, category-level analyses (especially those using the GoogleTagClassifier taxonomy) are based on a shorter interval beginning in May 2025, reflecting when consistent tagging became available on OpenRouter. In particular, detailed task classification fields (e.g., tags such as ",ex,", ",ey,", or ",(0,t.jsx)("em",{children:"Technology"}),") were only added in mid-2025. Consequently, all findings in the Categories section should be interpreted as representative of mid-2025 usage rather than the entire prior year."]}),(0,t.jsx)("p",{children:"Unless otherwise specified, all time-series aggregates are computed on a weekly basis using UTC-normalized timestamps, summing prompt and completion tokens. This approach ensures comparability across model families and minimizes bias from transient spikes or regional time-zone effects."})]})]}),ej[68]=eb):eb=ej[68],ej[69]!==V?(ev=(0,t.jsxs)(a.Section,{id:"data-and-methodology",children:[e,g,V,eu,ep,eb]}),ej[69]=V,ej[70]=ev):ev=ej[70],ev}],457357)},778769,e=>{"use strict";var t=e.i(222666),n=e.i(505962),o=e.i(553215),s=e.i(712621),r=e.i(252358),i=e.i(596694),a=e.i(831186);function l(e){let i,a,l=(0,n.c)(6),{url:c}=e,d=(0,r.usePosthogClient)();return l[0]!==d||l[1]!==c?(i=()=>{d.capture(o.PostHogEvent.ClickStateOfAIExternalLink,{url:c,link_text:c,section:"references"})},l[0]=d,l[1]=c,l[2]=i):i=l[2],l[3]!==i||l[4]!==c?(a=(0,t.jsx)(s.ExternalLink,{href:c,target:"_blank",rel:"noreferrer",className:"text-sm break-all focus-visible:border-focus-border focus-visible:shadow-focus focus-visible:outline-none",onClick:i,children:c}),l[3]=i,l[4]=c,l[5]=a):a=l[5],a}e.s(["References",0,function(){let e,o,s,r,c,d,m,h,u,f,p,g,x,y,b,v,j,_,w,S=(0,n.c)(19);return S[0]===Symbol.for("react.memo_cache_sentinel")?(e=(0,t.jsx)(i.Heading,{level:2,children:"References"}),S[0]=e):e=S[0],S[1]===Symbol.for("react.memo_cache_sentinel")?(o=(0,t.jsx)("em",{children:"arXiv preprint arXiv:2511.15080"}),S[1]=o):o=S[1],S[2]===Symbol.for("react.memo_cache_sentinel")?(s=(0,t.jsxs)("li",{className:"break-words mt-4 first:mt-0",children:["R. Appel, J. Zhao, C. Noll, O. K. Cheche, and W. E. Brown Jr. Anthropic economic index report: Uneven geographic and enterprise AI adoption."," ",o,", 2025. URL"," ",(0,t.jsx)(l,{url:"https://arxiv.org/abs/2511.15080"}),"."]}),S[2]=s):s=S[2],S[3]===Symbol.for("react.memo_cache_sentinel")?(r=(0,t.jsx)("em",{children:"NBER Working Paper 34255"}),S[3]=r):r=S[3],S[4]===Symbol.for("react.memo_cache_sentinel")?(c=(0,t.jsxs)("li",{className:"break-words mt-4 first:mt-0",children:["A. Chatterji, T. Cunningham, D. J. Deming, Z. Hitzig, C. Ong, C. Y. Shan, and K. Wadman. How people use chatgpt. ",r,", 2025. URL"," ",(0,t.jsx)(l,{url:"https://cdn.openai.com/pdf/a253471f-8260-40c6-a2cc-aa93fe9f142e/economic-research-chatgpt-usage-paper.pdf"}),"."]}),S[4]=c):c=S[4],S[5]===Symbol.for("react.memo_cache_sentinel")?(d=(0,t.jsx)("em",{children:"arXiv preprint arXiv:2405.01470"}),S[5]=d):d=S[5],S[6]===Symbol.for("react.memo_cache_sentinel")?(m=(0,t.jsxs)("li",{className:"break-words mt-4 first:mt-0",children:["W. Zhao, X. Ren, J. Hessel, C. Cardie, Y. Choi, and Y. Deng. WildChat: 1M ChatGPT interaction logs in the wild. ",d,", 2024. URL"," ",(0,t.jsx)(l,{url:"https://arxiv.org/abs/2405.01470"}),"."]}),S[6]=m):m=S[6],S[7]===Symbol.for("react.memo_cache_sentinel")?(h=(0,t.jsx)("em",{children:"arXiv preprint arXiv:2412.16720"}),S[7]=h):h=S[7],S[8]===Symbol.for("react.memo_cache_sentinel")?(u=(0,t.jsxs)("li",{className:"break-words mt-4 first:mt-0",children:["OpenAI. OpenAI o1 system card. ",h,", 2024. URL"," ",(0,t.jsx)(l,{url:"https://arxiv.org/abs/2412.16720"}),"."]}),S[8]=u):u=S[8],S[9]===Symbol.for("react.memo_cache_sentinel")?(f=(0,t.jsx)("em",{children:"arXiv preprint arXiv:2403.04132"}),S[9]=f):f=S[9],S[10]===Symbol.for("react.memo_cache_sentinel")?(p=(0,t.jsxs)("li",{className:"break-words mt-4 first:mt-0",children:["W. L. Chiang, L. Zheng, Y. Sheng, A. N. Angelopoulos, T. Li, D. Li, H. Zhang, B. Zhu, M. Jordan, J. Gonzalez, and I. Stoica. Chatbot Arena: An open platform for evaluating LLMs by human preference. ",f,", 2024. URL"," ",(0,t.jsx)(l,{url:"https://arxiv.org/abs/2403.04132"}),"."]}),S[10]=p):p=S[10],S[11]===Symbol.for("react.memo_cache_sentinel")?(g=(0,t.jsx)("em",{children:"Advances in Neural Information Processing Systems"}),S[11]=g):g=S[11],S[12]===Symbol.for("react.memo_cache_sentinel")?(x=(0,t.jsxs)("li",{className:"break-words mt-4 first:mt-0",children:["J. Wei, X. Wang, D. Schuurmans, M. Bosma, E. H. Chi, F. Xia, Q. Le, and D. Zhou. Chain-of-thought prompting elicits reasoning in large language models."," ",g,", 35:24824–24837, 2022. URL"," ",(0,t.jsx)(l,{url:"https://proceedings.neurips.cc/paper_files/paper/2022/hash/9d5609613524ecf4f15af0f7b31abca4-Abstract-Conference.html"}),"."]}),S[12]=x):x=S[12],S[13]===Symbol.for("react.memo_cache_sentinel")?(y=(0,t.jsx)("em",{children:"International Conference on Learning Representations (ICLR)"}),S[13]=y):y=S[13],S[14]===Symbol.for("react.memo_cache_sentinel")?(b=(0,t.jsxs)("li",{className:"break-words mt-4 first:mt-0",children:["S. Yao, J. Zhao, D. Yu, N. Du, I. Shafran, K. Narasimhan, and Y. Cao. ReAct: Synergizing reasoning and acting in language models."," ",y,", 2023. URL"," ",(0,t.jsx)(l,{url:"https://arxiv.org/abs/2210.03629"}),"."]}),S[14]=b):b=S[14],S[15]===Symbol.for("react.memo_cache_sentinel")?(v=(0,t.jsx)("em",{children:"arXiv preprint arXiv:2407.21783"}),S[15]=v):v=S[15],S[16]===Symbol.for("react.memo_cache_sentinel")?(j=(0,t.jsxs)("li",{className:"break-words mt-4 first:mt-0",children:["A. Grattafiori, A. Dubey, A. Jauhri, A. Pandey, A. Kadian, A. Al-Dahle, A. Letman, A. Mathur, A. Schelten, A. Yang, A. Fan, et al. The Llama 3 Herd of Models."," ",v,", 2024. URL"," ",(0,t.jsx)(l,{url:"https://arxiv.org/abs/2407.21783"}),"."]}),S[16]=j):j=S[16],S[17]===Symbol.for("react.memo_cache_sentinel")?(_=(0,t.jsx)("em",{children:"arXiv preprint arXiv:2412.19437"}),S[17]=_):_=S[17],S[18]===Symbol.for("react.memo_cache_sentinel")?(w=(0,t.jsxs)(a.Section,{id:"references",children:[e,(0,t.jsx)("div",{className:"prose max-w-none break-words",children:(0,t.jsxs)("ol",{className:"list-decimal pl-6 break-words",children:[s,c,m,u,p,x,b,j,(0,t.jsxs)("li",{className:"break-words mt-4 first:mt-0",children:["DeepSeek-AI, A. Liu, B. Feng, B. Xue, B. Wang, B. Wu, C. Lu, C. Zhao, C. Deng, C. Zhang, et al. DeepSeek-V3 technical report. ",_,", 2024. URL"," ",(0,t.jsx)(l,{url:"https://arxiv.org/abs/2412.19437"}),"."]})]})})]}),S[18]=w):w=S[18],w}])},155210,e=>{"use strict";var t=e.i(222666),n=e.i(505962),o=e.i(649188),s=e.i(553215),r=e.i(126154),i=e.i(712621),a=e.i(252358);e.s(["StateOfAIHero",0,function(){let e,l,c,d,m,h,u,f,p,g,x,y,b,v,j,_=(0,n.c)(17),w=(0,a.usePosthogClient)();return _[0]===Symbol.for("react.memo_cache_sentinel")?(e=(0,t.jsx)("h1",{className:"text-4xl font-semibold leading-tight md:text-5xl",children:"State of AI"}),l=(0,t.jsx)("p",{className:"text-2xl font-semibold leading-tight md:text-3xl",children:"An Empirical 100 Trillion Token Study with OpenRouter"}),_[0]=e,_[1]=l):(e=_[0],l=_[1]),_[2]===Symbol.for("react.memo_cache_sentinel")?(c=(0,t.jsxs)("div",{className:"flex flex-wrap items-center justify-center gap-x-4 gap-y-1",children:[(0,t.jsxs)("span",{children:[(0,t.jsx)("strong",{children:"Malika Aubakirova"}),(0,t.jsx)("sup",{className:"ml-1",children:"*"})]}),(0,t.jsxs)("span",{children:[(0,t.jsx)("strong",{children:"Alex Atallah"}),(0,t.jsx)("sup",{className:"ml-1",children:"†"})]}),(0,t.jsxs)("span",{children:[(0,t.jsx)("strong",{children:"Chris Clark"}),(0,t.jsx)("sup",{className:"ml-1",children:"†"})]}),(0,t.jsxs)("span",{children:[(0,t.jsx)("strong",{children:"Justin Summerville"}),(0,t.jsx)("sup",{className:"ml-1",children:"†"})]}),(0,t.jsxs)("span",{children:[(0,t.jsx)("strong",{children:"Anjney Midha"}),(0,t.jsx)("sup",{className:"ml-1",children:"*"})]})]}),_[2]=c):c=_[2],_[3]===Symbol.for("react.memo_cache_sentinel")?(d=(0,t.jsx)("sup",{children:"*"}),_[3]=d):d=_[3],_[4]===Symbol.for("react.memo_cache_sentinel")?(m=(0,t.jsx)(r.DotSeparator,{className:"mx-2"}),_[4]=m):m=_[4],_[5]===Symbol.for("react.memo_cache_sentinel")?(h=(0,t.jsxs)("div",{className:"text-xs text-muted-foreground",children:[d," a16z (Andreessen Horowitz) ",m," ",(0,t.jsx)("sup",{children:"†"})," ","OpenRouter Inc."]}),_[5]=h):h=_[5],_[6]===Symbol.for("react.memo_cache_sentinel")?(u=(0,t.jsx)("sup",{children:"*"}),_[6]=u):u=_[6],_[7]===Symbol.for("react.memo_cache_sentinel")?(f=(0,t.jsxs)("div",{className:"flex flex-col gap-4",children:[e,l,(0,t.jsxs)("div",{className:"flex flex-col gap-2 text-sm text-muted-foreground md:text-base",children:[c,h,(0,t.jsxs)("div",{className:"text-xs italic text-muted-foreground",children:[u," Lead contributors. Please see ",(0,t.jsx)("em",{children:"Contributions"})," section for details."]}),(0,t.jsx)("div",{className:"text-sm text-muted-foreground",children:"December, 2025"})]})]}),_[7]=f):f=_[7],_[8]===Symbol.for("react.memo_cache_sentinel")?(p=(0,t.jsx)("h2",{className:"text-xl font-semibold text-foreground",children:"Abstract"}),_[8]=p):p=_[8],_[9]===Symbol.for("react.memo_cache_sentinel")?(g=(0,t.jsx)("em",{children:"o1"}),_[9]=g):g=_[9],_[10]===Symbol.for("react.memo_cache_sentinel")?(x=(0,t.jsx)("em",{children:"foundational cohorts"}),_[10]=x):x=_[10],_[11]===Symbol.for("react.memo_cache_sentinel")?(y=(0,t.jsxs)("div",{className:"border-l-4 border-border pl-6 py-2",children:[p,(0,t.jsxs)("p",{className:"text-base leading-relaxed text-muted-foreground",children:["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,"," ",g,", 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 ",x,": early users whose engagement persists far longer than later cohorts. We term this phenomenon the Cinderella ",(0,t.jsx)("em",{children:'"Glass Slipper"'}),' 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.']})]}),_[11]=y):y=_[11],_[12]!==w?(b=()=>{w.capture(s.PostHogEvent.ClickStateOfAIPDFDownload)},_[12]=w,_[13]=b):b=_[13],_[14]===Symbol.for("react.memo_cache_sentinel")?(v=(0,t.jsx)(o.ArrowDownTrayIcon,{className:"h-4 w-4"}),_[14]=v):v=_[14],_[15]!==b?(j=(0,t.jsxs)("div",{className:"mb-16 flex w-full flex-col gap-8 text-center md:mb-24",children:[f,(0,t.jsxs)("div",{className:"prose mx-auto max-w-3xl text-left",children:[y,(0,t.jsx)("div",{className:"mt-4 flex justify-end lg:hidden",children:(0,t.jsxs)(i.ExternalLink,{href:"/assets/State-of-AI.pdf",download:"State-of-AI.pdf",className:"inline-flex items-center gap-2 text-sm font-medium text-foreground underline underline-offset-2 decoration-current/40 hover:decoration-current hover:text-accent-foreground",onClick:b,children:[v,"Download PDF"]})})]})]}),_[15]=b,_[16]=j):j=_[16],j}])},16735,e=>{"use strict";var t=e.i(222666),n=e.i(505962),o=e.i(553215),s=e.i(452473),r=e.i(252358);let i={anthropic2025:1,openai2025:2,zhao2024:3,openai2024o1:4,chiang2024:5,wei2022:6,yao2023:7,dubey2024:8,deepseek2024:9};function a(e,t){return e-t}function l(e){return void 0!==e}function c(e){return i[e]}e.s(["Citation",0,function(e){let i,d,m,h,u,f,p,g=(0,n.c)(16),{cite:x,className:y}=e,b=(0,r.usePosthogClient)();g[0]!==x?(i=Array.isArray(x)?x:[x],g[0]=x,g[1]=i):i=g[1];let v=i;if(g[2]!==v){h=Symbol.for("react.early_return_sentinel");e:{if(0===(d=v.map(c).filter(l).sort(a)).length){h=null;break e}m=d.join(", ")}g[2]=v,g[3]=d,g[4]=m,g[5]=h}else d=g[3],m=g[4],h=g[5];if(h!==Symbol.for("react.early_return_sentinel"))return h;let j=`[${m}]`;g[6]!==x||g[7]!==b||g[8]!==d?(u=()=>{b.capture(o.PostHogEvent.ClickStateOfAICitation,{citation_keys:Array.isArray(x)?x:[x],reference_numbers:d});let e=document.querySelector("#references");e&&(e.scrollIntoView({behavior:"auto",block:"start"}),window.history.pushState(null,"","#references"))},g[6]=x,g[7]=b,g[8]=d,g[9]=u):u=g[9];let _=u;return g[10]!==y?(f=(0,s.cn)("text-muted-foreground hover:underline cursor-pointer","focus-visible:border-focus-border focus-visible:shadow-focus focus-visible:outline-none",y),g[10]=y,g[11]=f):f=g[11],g[12]!==j||g[13]!==_||g[14]!==f?(p=(0,t.jsx)("sup",{children:(0,t.jsx)("button",{type:"button",className:f,onClick:_,children:j})}),g[12]=j,g[13]=_,g[14]=f,g[15]=p):p=g[15],p}])},596694,e=>{"use strict";var t=e.i(222666),n=e.i(505962),o=e.i(290861),s=e.i(553215),r=e.i(452473);e.i(810810);var i=e.i(702378),a=e.i(252358),l=e.i(781656),c=e.i(270401);e.s(["Heading",0,function(e){let d,m,h,u,f,p,g,x,y=(0,n.c)(24),{level:b,children:v,className:j}=e,_=(0,a.usePosthogClient)(),w="string"==typeof v?v:String(v);y[0]!==w?(d=(0,l.sanitizeForSelector)(w.replaceAll(/\s/g,"-")).toLowerCase(),y[0]=w,y[1]=d):d=y[1];let S=d,N=2===b?"h2":"h3";y[2]!==S?(m=()=>{if(window.location.hash===`#${S}`){let e=window.setTimeout(()=>{let e=document.querySelector(`#${S}`);e&&e.scrollIntoView({behavior:"auto",block:"start"})},100);return()=>window.clearTimeout(e)}},h=[S],y[2]=S,y[3]=m,y[4]=h):(m=y[3],h=y[4]),(0,c.useEffect)(m,h);let k=2===b&&"mb-6 text-3xl font-semibold",C=3===b&&"mb-4 text-2xl font-semibold";y[5]!==j||y[6]!==k||y[7]!==C?(u=(0,r.cn)("group relative flex items-center justify-start gap-2 scroll-mt-24",k,C,j),y[5]=j,y[6]=k,y[7]=C,y[8]=u):u=y[8];let A=`#${S}`;return y[9]!==S||y[10]!==b||y[11]!==_||y[12]!==w?(f=()=>{_.capture(s.PostHogEvent.ClickStateOfAIHeadingLink,{heading_id:S,heading_text:w,heading_level:b}),window.location.hash=S;let e=document.querySelector(`#${S}`);e&&e.scrollIntoView({behavior:"smooth",block:"start"})},y[9]=S,y[10]=b,y[11]=_,y[12]=w,y[13]=f):f=y[13],y[14]===Symbol.for("react.memo_cache_sentinel")?(p=(0,t.jsx)(o.LinkIcon,{className:"size-3"}),y[14]=p):p=y[14],y[15]!==A||y[16]!==f?(g=(0,t.jsx)(i.CopyToClipboardButton,{hideCopyText:!0,prependUrl:!0,text:A,onClick:f,icon:p,className:"opacity-0 transition-opacity duration-200 group-hover:opacity-100 focus-visible:border-focus-border focus-visible:shadow-focus focus-visible:outline-none"}),y[15]=A,y[16]=f,y[17]=g):g=y[17],y[18]!==N||y[19]!==v||y[20]!==S||y[21]!==g||y[22]!==u?(x=(0,t.jsxs)(N,{id:S,className:u,children:[v,g]}),y[18]=N,y[19]=v,y[20]=S,y[21]=g,y[22]=u,y[23]=x):x=y[23],x}])},923786,e=>{"use strict";var t=e.i(222666),n=e.i(505962),o=e.i(551001),s=e.i(553215),r=e.i(452473),i=e.i(763392),a=e.i(883992),l=e.i(528164),c=e.i(778084),d=e.i(394333),m=e.i(691038),h=e.i(252358),u=e.i(779100),f=e.i(370066),p=e.i(706527),g=e.i(179273),x=e.i(270401),y=e.i(690387);e.i(406201);var b=e.i(806024),v=e.i(251142);let j={p:e=>(0,t.jsx)(a.MarkdownP,{...e,className:(0,r.cn)("leading-relaxed text-sm",e.className)})};async function _(e){try{let t=await fetch(`/api/frontend/v1/models/search?q=${encodeURIComponent(e)}`);if(!t.ok)return null;return(await t.json()).model||null}catch{return null}}function w(e){let r,a,c,d,m,x,_,w,S,N,k,C,A,L,T,I,R,P,M,E,O,D,H,z,W=(0,n.c)(66),{model:U}=e,F=(0,u.useLocale)();W[0]!==U.endpoint?.display_pricing?(r=U.endpoint?.display_pricing??[],W[0]=U.endpoint?.display_pricing,W[1]=r):r=W[1];let B=r;W[2]!==B||W[3]!==F||W[4]!==U.endpoint?(a=(0,y.deriveEndpointInlinePricing)(F,U.endpoint,B),W[2]=B,W[3]=F,W[4]=U.endpoint,W[5]=a):a=W[5];let G=a;W[6]!==B||W[7]!==U.endpoint?(c=(0,y.isModelFree)(U.endpoint,B),W[6]=B,W[7]=U.endpoint,W[8]=c):c=W[8];let q=c,X=U.context_length;W[9]!==U?(d=(0,p.getModelVariantSlug)(U),W[9]=U,W[10]=d):d=W[10];let Y=d,$=(0,h.usePosthogClient)();W[11]!==U.name||W[12]!==Y||W[13]!==$?(m=e=>{$.capture(s.PostHogEvent.ClickStateOfAIModelLink,{model_slug:Y,model_name:U.name,link_type:e})},W[11]=U.name,W[12]=Y,W[13]=$,W[14]=m):m=W[14];let J=m,K=`/${Y}`;W[15]!==J?(x=e=>{e.stopPropagation(),J("name")},W[15]=J,W[16]=x):x=W[16],W[17]!==U.name?(_=(0,t.jsx)("h3",{className:"font-semibold text-base text-foreground group-hover/link:underline break-words",children:U.name}),W[17]=U.name,W[18]=_):_=W[18];let V=U.endpoint?.variant??g.Variant.Standard;W[19]!==V?(w=(0,t.jsx)(v.VariantIcon,{variant:V,className:"size-4 shrink-0"}),W[19]=V,W[20]=w):w=W[20],W[21]!==w||W[22]!==K||W[23]!==x||W[24]!==_?(S=(0,t.jsxs)(b.Link,{href:K,className:"flex items-center gap-2 group/link",onClick:x,children:[_,w]}),W[21]=w,W[22]=K,W[23]=x,W[24]=_,W[25]=S):S=W[25];let Z=`/${Y}`;W[26]!==J?(N=e=>{e.stopPropagation(),J("slug")},W[26]=J,W[27]=N):N=W[27],W[28]!==Y||W[29]!==Z||W[30]!==N?(k=(0,t.jsx)(b.Link,{href:Z,className:"text-overline text-muted-foreground hover:underline break-all",onClick:N,children:Y}),W[28]=Y,W[29]=Z,W[30]=N,W[31]=k):k=W[31],W[32]!==S||W[33]!==k?(C=(0,t.jsx)("div",{className:"flex items-start justify-between gap-2",children:(0,t.jsxs)("div",{className:"flex-1 min-w-0",children:[S,k]})}),W[32]=S,W[33]=k,W[34]=C):C=W[34],W[35]!==U.description?(A=U.description?.replaceAll("\n"," "),W[35]=U.description,W[36]=A):A=W[36],W[37]!==A?(L=(0,t.jsx)(i.MarkdownContainer,{disableLink:!0,className:"text-sm w-auto line-clamp-4 break-words",components:j,children:A}),W[37]=A,W[38]=L):L=W[38],W[39]!==X||W[40]!==F?(T=(0,f.formatNumber)(X,{locale:F,preset:"compact-rounded-thousands"})??"—",W[39]=X,W[40]=F,W[41]=T):T=W[41],W[42]!==T?(I=(0,t.jsxs)("span",{children:[T," ","context"]}),W[42]=T,W[43]=I):I=W[43],W[44]!==q?(R=q&&(0,t.jsx)("span",{children:"Free"}),W[44]=q,W[45]=R):R=W[45],W[46]!==G||W[47]!==q?(P=!q&&G&&(0,t.jsxs)(t.Fragment,{children:[(0,t.jsx)("span",{className:"border-l"}),"custom"===G.kind?(0,t.jsx)("span",{children:G.label}):(0,t.jsxs)(t.Fragment,{children:[(0,t.jsxs)("span",{children:[G.inputPrice,G.inputUnit]}),(0,t.jsx)("span",{className:"border-l"}),(0,t.jsxs)("span",{children:[G.outputPrice,G.outputUnit]})]})]}),W[46]=G,W[47]=q,W[48]=P):P=W[48],W[49]!==I||W[50]!==R||W[51]!==P?(M=(0,t.jsxs)("div",{className:"flex flex-wrap gap-x-3 gap-y-1 text-xs text-muted-foreground",children:[I,R,P]}),W[49]=I,W[50]=R,W[51]=P,W[52]=M):M=W[52],W[53]!==J?(E=e=>{e.stopPropagation(),J("button")},W[53]=J,W[54]=E):E=W[54];let Q=`/${Y}`;return W[55]===Symbol.for("react.memo_cache_sentinel")?(O=(0,t.jsx)(o.ArrowRightIcon,{className:"size-3"}),W[55]=O):O=W[55],W[56]!==Q?(D=(0,t.jsxs)(b.Link,{href:Q,target:"_blank",rel:"noreferrer",children:["View model",O]}),W[56]=Q,W[57]=D):D=W[57],W[58]!==E||W[59]!==D?(H=(0,t.jsx)(l.Button,{variant:"outline",size:"sm",className:"w-full",onClick:E,render:D}),W[58]=E,W[59]=D,W[60]=H):H=W[60],W[61]!==C||W[62]!==L||W[63]!==M||W[64]!==H?(z=(0,t.jsxs)("div",{className:"flex flex-col gap-3 min-w-80 max-w-md",children:[C,L,M,H]}),W[61]=C,W[62]=L,W[63]=M,W[64]=H,W[65]=z):z=W[65],z}e.s(["ModelHoverPopover",0,function(e){let o,i,a,l,u,f,p,g,y,b,v,j,S,N,k,C,A=(0,n.c)(41),{modelSlug:L,children:T,className:I}=e,[R,P]=(0,x.useState)(!1),[M,E]=(0,x.useState)(null),[O,D]=(0,x.useState)(!1),[H,z]=(0,x.useState)(!1),W=(0,m.useMatchMedia)("(hover: hover) and (pointer: fine)"),U=x.default.useRef(null),F=x.default.useRef(!1),B=(0,h.usePosthogClient)();A[0]!==M||A[1]!==L||A[2]!==R||A[3]!==B?(o=()=>{if(!R||M)return;F.current||(B.capture(s.PostHogEvent.HoverStateOfAIModelPopover,{model_slug:L}),F.current=!0);let e=!1;return D(!0),z(!1),_(L).then(t=>{!e&&(E(t),D(!1),t||z(!0))}).catch(()=>{e||(D(!1),z(!0))}),()=>{e=!0}},i=[R,L,M,B],A[0]=M,A[1]=L,A[2]=R,A[3]=B,A[4]=o,A[5]=i):(o=A[4],i=A[5]),(0,x.useEffect)(o,i),A[6]!==R?(a=()=>{if(!R){let e=setTimeout(()=>{E(null),z(!1),F.current=!1},200);return()=>clearTimeout(e)}},l=[R],A[6]=R,A[7]=a,A[8]=l):(a=A[7],l=A[8]),(0,x.useEffect)(a,l),A[9]===Symbol.for("react.memo_cache_sentinel")?(u=()=>()=>{null!==U.current&&window.clearTimeout(U.current)},f=[],A[9]=u,A[10]=f):(u=A[9],f=A[10]),(0,x.useEffect)(u,f),A[11]!==W?(p=()=>{W&&(null!==U.current&&(window.clearTimeout(U.current),U.current=null),P(!0))},A[11]=W,A[12]=p):p=A[12];let G=p;A[13]!==W?(g=()=>{W&&(U.current=window.setTimeout(()=>{P(!1),U.current=null},100))},A[13]=W,A[14]=g):g=A[14];let q=g;return A[15]!==I?(y=(0,r.cn)("cursor-pointer underline decoration-dotted underline-offset-2",I),A[15]=I,A[16]=y):y=A[16],A[17]!==T||A[18]!==y?(b=(0,t.jsx)("span",{className:y,children:T}),A[17]=T,A[18]=y,A[19]=b):b=A[19],A[20]!==G||A[21]!==q||A[22]!==b?(v=(0,t.jsx)(c.PopoverTrigger,{onMouseEnter:G,onMouseLeave:q,render:b}),A[20]=G,A[21]=q,A[22]=b,A[23]=v):v=A[23],A[24]!==O?(j=O&&(0,t.jsxs)("div",{className:"flex flex-col gap-3 min-w-80",children:[(0,t.jsx)(d.Skeleton,{className:"h-4 w-3/4"}),(0,t.jsx)(d.Skeleton,{className:"h-3 w-full"}),(0,t.jsx)(d.Skeleton,{className:"h-3 w-5/6"}),(0,t.jsx)(d.Skeleton,{className:"h-3 w-4/6"})]}),A[24]=O,A[25]=j):j=A[25],A[26]!==H||A[27]!==L?(S=H&&(0,t.jsxs)("div",{className:"text-sm text-muted-foreground",children:['Model "',L,'" not found']}),A[26]=H,A[27]=L,A[28]=S):S=A[28],A[29]!==M?(N=M&&(0,t.jsx)(w,{model:M}),A[29]=M,A[30]=N):N=A[30],A[31]!==G||A[32]!==q||A[33]!==j||A[34]!==S||A[35]!==N?(k=(0,t.jsxs)(c.PopoverContent,{onMouseEnter:G,onMouseLeave:q,side:"top",align:"start",sideOffset:8,className:"p-4 !w-auto max-w-[90vw]",children:[j,S,N]}),A[31]=G,A[32]=q,A[33]=j,A[34]=S,A[35]=N,A[36]=k):k=A[36],A[37]!==R||A[38]!==v||A[39]!==k?(C=(0,t.jsxs)(c.Popover,{open:R,onOpenChange:P,children:[v,k]}),A[37]=R,A[38]=v,A[39]=k,A[40]=C):C=A[40],C},"ModelPreview",0,w])},831186,e=>{"use strict";var t=e.i(222666),n=e.i(505962);e.s(["Section",0,function(e){let o,s=(0,n.c)(4),{id:r,children:i,className:a}=e,l=`prose max-w-none my-12 first:mt-0 scroll-mt-24 prose-headings:font-semibold prose-p:text-prose prose-p:leading-prose prose-p:text-prose-body prose-ul:my-4 prose-li:my-1 prose-code:before:content-none prose-code:after:content-none ${a??""}`;return s[0]!==i||s[1]!==r||s[2]!==l?(o=(0,t.jsx)("section",{id:r,className:l,children:i}),s[0]=i,s[1]=r,s[2]=l,s[3]=o):o=s[3],o}])},27819,e=>{"use strict";var t=e.i(222666),n=e.i(649188),o=e.i(553215),s=e.i(452473),r=e.i(712621),i=e.i(980351),a=e.i(252358),l=e.i(781656),c=e.i(270401);function d({title:e,level:t,children:n}){return{id:(0,l.sanitizeForSelector)(e.replaceAll(/\s/g,"-")).toLowerCase(),title:e,level:t,children:n}}let m=[d({title:"Introduction",level:2}),d({title:"Data and Methodology",level:2,children:[d({title:"OpenRouter Platform and Dataset",level:3}),d({title:"GoogleTagClassifier for Content Categorization",level:3}),d({title:"Model and Token Variants",level:3}),d({title:"Geographic Segmentation",level:3}),d({title:"Time Frame and Coverage",level:3})]}),d({title:"Open vs. Closed Source Models",level:2,children:[d({title:"Key Open Source Players",level:3}),d({title:"The Model Size vs. Market Fit: Medium is the New Small",level:3}),d({title:"What Are Open Source Models Used For?",level:3})]}),d({title:"The Rise of Agentic Inference",level:2,children:[d({title:"Reasoning Models Now Represent Half of All Usage",level:3}),d({title:"Rising Adoption of Tool-Calling",level:3}),d({title:"The Anatomy of Prompt-Completion Shapes",level:3}),d({title:"Longer Sequences, More Complex Interactions",level:3}),d({title:"Implications: Agentic Inference Is the New Default",level:3})]}),d({title:"Categories: How Are People Using LLMs?",level:2,children:[d({title:"Dominant Categories",level:3}),d({title:"Tag Composition Within Categories",level:3}),d({title:"Author-Level Insights by Category",level:3})]}),d({title:"Geography: How LLM Usage Differs Across Regions",level:2,children:[d({title:"Regional Distribution of Usage",level:3}),d({title:"Language Distribution",level:3})]}),d({title:"Analysis of LLM User Retention",level:2,children:[d({title:'The Cinderella "Glass Slipper" Phenomenon',level:3})]}),d({title:"Cost vs. Usage Dynamics",level:2,children:[d({title:"Analysis of AI Workload Segmentation by Categories",level:3}),d({title:"Effective Cost vs Usage of AI Models",level:3})]}),d({title:"Discussion",level:2}),d({title:"Limitations",level:2}),d({title:"Conclusion",level:2}),d({title:"References",level:2}),d({title:"Contributions",level:2}),d({title:"Appendix",level:2})];e.s(["TableOfContents",0,function(){let[e,l]=(0,c.useState)(null),d=(0,a.usePosthogClient)(),h=(0,c.useRef)(new Set),u=(0,i.useHydrated)(),f=t=>t.id===e||!!t.children&&t.children.some(e=>f(e));(0,c.useLayoutEffect)(()=>{if(!u)return;let e=window.location.hash;if(!e)return;let t=e.slice(1),n=0,o=()=>{n++;let e=document.getElementById(t);if(e){let t=e.getBoundingClientRect().top+window.pageYOffset-100;window.scrollTo({top:t,behavior:"auto"})}else n<10&&requestAnimationFrame(o)};requestAnimationFrame(o)},[u]),(0,c.useEffect)(()=>{let e=(function e(t){let n=[];for(let o of t)n.push(o),o.children&&n.push(...e(o.children));return n})(m).map(e=>document.getElementById(e.id)).filter(e=>null!==e);if(0===e.length)return;let t=null,n=window.scrollY,s=()=>{null===t&&(t=requestAnimationFrame(()=>{let s=window.scrollY;if(10>Math.abs(s-n)){t=null;return}n=s;let r=s+100,i=window.innerHeight,a=document.documentElement.scrollHeight,c=Math.round((s+i)/a*100);for(let e of[25,50,75,100])c>=e&&!h.current.has(e)&&(h.current.add(e),d.capture(o.PostHogEvent.StateOfAIScrollDepth,{scroll_depth_percent:e}));let m=null;if(a-(s+i)<200){let t=e.at(-1);t&&(m=t.id)}else{for(let t=e.length-1;t>=0;t--){let n=e[t];if(n&&n.offsetTop<=r){m=n.id;break}}let t=e[0];t&&r<t.offsetTop&&(m=t.id)}l(m),t=null}))};return s(),window.addEventListener("scroll",s,{passive:!0}),()=>{window.removeEventListener("scroll",s),null!==t&&cancelAnimationFrame(t)}},[]);let p=(n,r=0)=>{let i=e===n.id,a=n.children&&n.children.length>0&&f(n);return(0,t.jsxs)("li",{className:(0,s.cn)("list-none",r>0&&"ml-4","mt-0.5 first:mt-0"),children:[(0,t.jsx)("a",{href:`#${n.id}`,className:(0,s.cn)("block py-1.5 text-sm transition-colors","hover:text-foreground","focus-visible:border-focus-border focus-visible:shadow-focus focus-visible:outline-none",0===r?"font-medium":"font-normal",i?"text-foreground":"text-muted-foreground",0===r&&"mt-2 first:mt-0"),onClick:e=>{e.preventDefault();let t=document.getElementById(n.id);if(t){let e=t.getBoundingClientRect().top+window.scrollY-80;d.capture(o.PostHogEvent.ClickTableOfContentsLink,{section_id:n.id,section_title:n.title,section_level:n.level}),window.scrollTo({top:e,behavior:"auto"})}},children:n.title}),a&&n.children&&(0,t.jsx)("ul",{className:"mt-1",children:n.children.map(e=>p(e,r+1))})]},n.id)};return(0,t.jsx)("nav",{className:(0,s.cn)("hidden lg:block","sticky top-28","self-start","w-64 shrink-0"),"aria-label":"Table of contents",children:(0,t.jsxs)("div",{className:"border-r border-border lg:pr-8 md:pr-6",children:[(0,t.jsxs)("div",{className:"mb-4 flex items-center justify-between",children:[(0,t.jsx)("h2",{className:"text-base font-semibold text-foreground",children:"Contents"}),(0,t.jsxs)(r.ExternalLink,{href:"/assets/State-of-AI.pdf",download:"State-of-AI.pdf",className:"inline-flex items-center gap-1.5 text-xs font-medium text-foreground underline underline-offset-2 decoration-current/40 hover:decoration-current hover:text-accent-foreground",onClick:()=>{d.capture(o.PostHogEvent.ClickStateOfAIPDFDownload)},children:[(0,t.jsx)(n.ArrowDownTrayIcon,{className:"h-3.5 w-3.5"}),"PDF"]})]}),(0,t.jsx)("ul",{children:m.map(e=>p(e))})]})})}])}]);

//# sourceMappingURL=0sd-mow04csbn.js.map
//# chunkId=01a0879f-7ac3-7280-89a6-dfff314ac3f8