https://aiagentstore.ai/_next/static/chunks/pages/ai-models/benchmarks-21078d2812f5d4ec.js

1(self.webpackChunk_N_E=self.webpackChunk_N_E||[]).push([[1444],{9449:(e,t,n)=>{"use strict";n.r(t),n.d(t,{default:()=>i});var a=n(69519),s=n(40927),r=n(17528);function i(){return(0,a.jsxs)(s.Ay,{title:"AI Benchmarks Explained: LiveBench, Arena, Tool Calling & LLM Metrics",description:"Understand AI model benchmarks in plain English. Learn what coding, reasoning, instruction following, Arena and BFCL scores measure—and what they miss.",path:"/ai-models/benchmarks",breadcrumbs:[{name:"Benchmarks explained",item:"https://aiagentstore.ai/ai-models/benchmarks"}],children:[(0,a.jsx)(s.QE,{title:"AI model benchmarks explained",description:"A field guide to LiveBench, Arena, tool-use and document-grounding benchmarks, with source configurations and limitations.",path:"/ai-models/benchmarks",citations:r.Hl.map(e=>e.url)}),(0,a.jsxs)("section",{className:s.R7.articleHero,children:[(0,a.jsx)("p",{className:s.R7.eyebrow,children:"A FIELD GUIDE TO THE NUMBERS"}),(0,a.jsxs)("h1",{children:["Less benchmark jargon.",(0,a.jsx)("br",{}),(0,a.jsx)("em",{children:"More useful answers."})]}),(0,a.jsx)("p",{children:"A benchmark asks a model to do a particular kind of work. Knowing what the test asks is more useful than knowing who came first."}),(0,a.jsx)(s.OL,{})]}),(0,a.jsx)("div",{className:s.R7.benchmarkGrid,children:Object.entries(r.Cb).map(e=>{let[t,n]=e;return(0,a.jsxs)("article",{id:t.toLowerCase().replaceAll(" ","-"),children:[(0,a.jsx)("span",{className:s.R7.kicker,children:["fit","value","consensus"].includes(t)?"AI AGENT STORE RATING":"MEASURE EXPLAINED"}),(0,a.jsx)("h2",{children:n.name}),(0,a.jsx)("p",{children:n.text})]},t)})}),(0,a.jsxs)("section",{className:s.R7.referenceSection,children:[(0,a.jsx)("p",{className:s.R7.eyebrow,children:"A DIFFERENT TEST. A DIFFERENT SNAPSHOT."}),(0,a.jsx)("h2",{children:"Tool-use evidence: Berkeley BFCL V4"}),(0,a.jsx)("p",{children:"These published configurations are from an older, separate evaluation, last updated 12 April 2026. They are useful context for understanding tool use; they do not measure the latest versions in our model table and are excluded from our aggregate ratings. FC means native function calling; Prompt means text-based prompting."}),(0,a.jsx)("div",{className:s.R7.tableScroll,children:(0,a.jsxs)("table",{className:s.R7.compareTable,children:[(0,a.jsxs)("caption",{children:["Published BFCL V4 results \xb7 ",(0,a.jsx)("a",{href:"https://gorilla.cs.berkeley.edu/leaderboard.html",children:"Source and methodology ↗"})]}),(0,a.jsx)("thead",{children:(0,a.jsxs)("tr",{children:[(0,a.jsx)("th",{children:"Tested configuration"}),(0,a.jsxs)("th",{children:["Overall accuracy ",(0,a.jsx)(s._V,{metric:"bfcl"})]}),(0,a.jsx)("th",{children:"Multi-turn"}),(0,a.jsx)("th",{children:"Memory"})]})}),(0,a.jsx)("tbody",{children:r.vN.toolUseReference.map(e=>(0,a.jsxs)("tr",{children:[(0,a.jsx)("th",{children:e.model}),(0,a.jsxs)("td",{children:[(0,r.ai)(e.accuracy,2),"%"]}),(0,a.jsxs)("td",{children:[(0,r.ai)(e.multiTurn,2),"%"]}),(0,a.jsxs)("td",{children:[(0,r.ai)(e.memory,2),"%"]})]},e.model))})]})})]}),(0,a.jsxs)("section",{className:s.R7.referenceSection,children:[(0,a.jsx)("p",{className:s.R7.eyebrow,children:"GROUNDED ANSWERS NEED THEIR OWN TEST"}),(0,a.jsx)("h2",{children:"Hallucination is not one universal number."}),(0,a.jsxs)("p",{children:["Vectara’s document-summary evaluation checks whether an answer stays supported by the supplied document. The ",r.vN.groundingDate," snapshot below uses Vectara’s exact named configurations, not the reasoning configurations in our LiveBench table. GPT-6.1 Sol and Claude Sonnet 5.5 do not have verified matching observations here."]}),(0,a.jsx)("div",{className:s.R7.tableScroll,children:(0,a.jsxs)("table",{className:s.R7.compareTable,children:[(0,a.jsxs)("caption",{children:["Document-summary reference \xb7 observed ",r.vN.groundingDate]}),(0,a.jsx)("thead",{children:(0,a.jsxs)("tr",{children:[(0,a.jsx)("th",{children:"Exact tested model"}),(0,a.jsx)("th",{children:"Hallucination rate"}),(0,a.jsx)("th",{children:"Factual consistency"}),(0,a.jsx)("th",{children:"Answer rate"})]})}),(0,a.jsx)("tbody",{children:r.vN.groundingReference.map(e=>(0,a.jsxs)("tr",{children:[(0,a.jsx)("th",{scope:"row",children:(0,a.jsx)("code",{children:e.model})}),(0,a.jsxs)("td",{children:[(0,r.ai)(e.hallucination,1),"%"]}),(0,a.jsxs)("td",{children:[(0,r.ai)(e.factualConsistency,1),"%"]}),(0,a.jsxs)("td",{children:[(0,r.ai)(e.answerRate,1),"%"]})]},e.model))})]})}),(0,a.jsxs)("p",{children:["These rates do not predict factual accuracy on open-ended questions or autonomous tasks. Answer rate and the evaluation judge also matter. We therefore keep them out of our scores. ",(0,a.jsx)("a",{href:"https://github.com/vectara/hallucination-leaderboard",children:"Read the dataset, configurations and limitations ↗"})]})]}),(0,a.jsxs)("section",{className:s.R7.cta,children:[(0,a.jsxs)("div",{children:[(0,a.jsx)("h2",{children:"Now put the evidence to work."}),(0,a.jsx)("p",{children:"Compare models with explanations beside every metric."})]}),(0,a.jsx)("a",{className:s.R7.primaryButton,href:"/ai-models#explorer",children:"Explore the model comparison ↗"})]})]})}},13153:(e,t,n)=>{(window.__NEXT_P=window.__NEXT_P||[]).push(["/ai-models/benchmarks",function(){return n(9449)}])},40927:(e,t,n)=>{"use strict";n.d(t,{Ay:()=>w,OL:()=>y,QE:()=>j,R7:()=>h.a,Sj:()=>u,_V:()=>m,cl:()=>f,e8:()=>b});var a=n(69519),s=n(86603),r=n.n(s),i=n(16879),o=n(31434),l=n(26514),d=n(17528),c=n(50791),h=n.n(c);function m(e){let{metric:t,label:n,description:s}=e,r=d.Cb[t],l=(0,i.useId)(),c=(0,i.useRef)(null),m=(0,i.useRef)(!1),[u,p]=(0,i.useState)(!1),[x,g]=(0,i.useState)({}),j=()=>

1{let e=c.current?.getBoundingClientRect();if(!e)return;let t=Math.min(340,window.innerWidth-32);g(window.innerWidth<=760?{left:16,right:16,bottom:92}:{left:Math.max(16,Math.min(window.innerWidth-t-16,e.left-t/2+e.width/2)),width:t,...e.top>190?{bottom:window.innerHeight-e.top+8}:{top:e.bottom+8}}),p(!0)},f=()=>{m.current=!1,p(!1)};return(0,i.useEffect)(()=>{if(!u)return;let e=e=>{c.current?.contains(e.target)||f()},t=e=>{"Escape"===e.key&&f()},n=()=>{m.current||document.activeElement!==c.current?f():j()};return document.addEventListener("pointerdown",e),document.addEventListener("keydown",t),window.addEventListener("scroll",n,!0),window.addEventListener("resize",f),()=>{document.removeEventListener("pointerdown",e),document.removeEventListener("keydown",t),window.removeEventListener("scroll",n,!0),window.removeEventListener("resize",f)}},[u]),(0,a.jsxs)("span",{className:h().helpWrap,children:[(0,a.jsx)("button",{ref:c,className:h().help,type:"button","aria-label":`Explain ${n||r?.name||t}`,"aria-describedby":u?l:void 0,"aria-expanded":u,onPointerEnter:e=>{"mouse"===e.pointerType&&j()},onPointerLeave:()=>{m.current||p(!1)},onFocus:j,onBlur:f,onClick:()=>{m.current?f():(m.current=!0,j())},children:"?"}),u&&(0,o.createPortal)((0,a.jsxs)("span",{id:l,role:"tooltip",className:h().instantTooltip,style:x,children:[(0,a.jsx)("strong",{children:r?.name||n}),s||r?.text||n]}),document.body)]})}function u(e){let{data:t}=e;return(0,a.jsx)("script",{type:"application/ld+json",dangerouslySetInnerHTML:{__html:JSON.stringify(t).replace(/</g,"\\u003c")}})}let p="https://aiagentstore.ai",x=`${p}/ai-models-social.png`,g={"@type":"Organization","@id":`${p}/#organization`,name:"AI Agent Store",url:p,logo:{"@type":"ImageObject",url:`${p}/logo-transparent.svg`}};function j(e){let{title:t,description:n,path:s,modified:r=d.vN.updated,citations:i=[]}=e,o=p+s;return(0,a.jsx)(u,{data:{"@context":"https://schema.org","@type":"TechArticle","@id":`${o}#article`,url:o,headline:t,description:n,dateModified:r,inLanguage:"en",image:x,author:g,publisher:g,mainEntityOfPage:{"@id":`${o}#webpage`},...i.length?{citation:i}:{}}})}function f(e){let{items:t,path:n}=e;return(0,a.jsx)(u,{data:{"@context":"https://schema.org","@type":"FAQPage",...n?{"@id":`${p}${n}#faq`}:{},mainEntity:t.map(e=>{let[t,n]=e;

1return{"@type":"Question",name:t,acceptedAnswer:{"@type":"Answer",text:n}}})}})}function b(e){let{name:t,description:n,path:s,modified:r,download:i,format:o="application/json",citations:l=[],variables:d=[]}=e,c=p+s;return(0,a.jsx)(u,{data:{"@context":"https://schema.org","@type":"Dataset","@id":`${c}#dataset-${i.split("/").pop()}`,name:t,description:n,url:c,dateModified:r,creator:g,isAccessibleForFree:!0,inLanguage:"en",isPartOf:{"@id":`${c}#webpage`},...l.length?{citation:l,isBasedOn:l}:{},...d.length?{variableMeasured:d.map(e=>({"@type":"PropertyValue",name:e}))}:{},distribution:{"@type":"DataDownload",contentUrl:p+i,encodingFormat:o}}})}function w(e){let{title:t,description:n,path:s="/ai-models",modified:i=d.vN.updated,children:o,breadcrumbs:c=[]}=e,m="https://aiagentstore.ai"+s,j=s.startsWith("/ai-models/reasoning-effort")?"/ai-models/reasoning-effort/methodology":"/ai-models/methodology",f=[{name:"Home",item:"https://aiagentstore.ai/"},{name:"AI models",item:"https://aiagentstore.ai/ai-models"},...c];return(0,a.jsxs)(a.Fragment,{children:[(0,a.jsxs)(r(),{children:[(0,a.jsx)("title",{children:`${t} | AI Agent Store`}),(0,a.jsx)("meta",{name:"description",content:n}),(0,a.jsx)("link",{rel:"canonical",href:m}),(0,a.jsx)("meta",{name:"robots",content:"index,follow,max-image-preview:large"}),(0,a.jsx)("meta",{property:"og:title",content:t}),(0,a.jsx)("meta",{property:"og:description",content:n}),(0,a.jsx)("meta",{property:"og:url",content:m}),(0,a.jsx)("meta",{property:"og:type",content:"website"}),(0,a.jsx)("meta",{property:"og:site_name",content:"AI Agent Store"}),(0,a.jsx)("meta",{property:"og:image",content:"https://aiagentstore.ai/ai-models-social.png"}),(0,a.jsx)("meta",{name:"twitter:card",content:"summary_large_image"}),(0,a.jsx)("meta",{name:"twitter:title",content:t}),(0,a.jsx)("meta",{name:"twitter:description",content:n}),(0,a.jsx)("meta",{name:"twitter:image",content:"https://aiagentstore.ai/ai-models-social.png"})]}),(0,a.jsxs)(r(),{children:[(0,a.jsx)("meta",{property:"og:locale",content:"en_US"}),(0,a.jsx)("meta",{property:"og:image:alt",content:"AI Agent Store model intelligence: benchmarks, pricing and reasoning effort"}),(0,a.jsx)("meta",{name:"twitter:image:alt",content:"AI Agent Store model intelligence: benchmarks, pricing and reasoning effort"})]}),(0,a.jsx)(u,{data:{"@context":"https://schema.org","@type":"WebPage","@id":`${m}#webpage`,url:m,name:t,description:n,dateModified:i,inLanguage:"en",publisher:g,isPartOf:{"@type":"WebSite","@id":`${p}/#website`,name:"AI Agent Store",url:p},primaryImageOfPage:{"@type":"ImageObject",url:x},breadcrumb:{"@id":`${m}#breadcrumb`}}}),(0,a.jsx)(u,{data:{"@context":"https://schema.org","@type":"BreadcrumbList","@id":`${m}#breadcrumb`,itemListElement:f.map((e,t)=>({"@type":"ListItem",position:t+1,...e}))}}),(0,a.jsxs)("div",{className:h().page,children:[(0,a.jsx)("a",{className:h().skip,href:"#model-main",children:"Skip to model comparison"}),(0,a.jsxs)("header",{className:h().header,children:[(0,a.jsxs)("a",{className:h().brand,href:"/",children:[(0,a.jsx)("img",{src:"/logo-transparent.svg",width:"28",height:"32",alt:""}),"AI Agent Store",(0,a.jsx)("span",{className:h().brandDivider,children:"/"}),(0,a.jsx)("span",{children:"Model intelligence"})]}),(0,a.jsxs)("nav",{"aria-label":"Model navigation",children:[(0,a.jsx)("a",{href:"/ai-models","aria-current":"/ai-models"===s?"page":void 0,children:"Explore models"}),(0,a.jsx)("a",{href:"/ai-models/reasoning-effort","aria-current":s.startsWith("/ai-models/reasoning-effort")?"page":void 0,children:"Reasoning effort"}),(0,a.jsx)("a",{href:"/ai-models/benchmarks",children:"Benchmarks explained"}),(0,a.jsx)("a",{href:j,children:"Our methodology"})]}),(0,a.jsx)("a",{className:h().headerCta,href:"/agent-teams",children:"Build an AI worker ↗"})]}),(0,a.jsxs)("main",{id:"model-main",className:h().main,children:["/ai-models"!==s&&(0,a.jsxs)("nav",{className:h().breadcrumb,"aria-label":"Breadcrumb",children:[(0,a.jsx)("a",{href:"/ai-models",children:"AI models"}),(0,a.jsx)("span",{children:"/"}),(0,a.jsx)("span",{children:c[c.length-1]?.name||t})]}),o]}),(0,a.jsxs)("div",{className:h().bottomNote,children:["AI Agent Store \xb7 Model intelligence \xb7 Snapshot ",d.vN.updated,(0,a.jsx)("a",{href:`${j}#sources`,children:"Sources, limitations & corrections ↗"})]})]}),(0,a.jsx)(l.default,{})]})}function y(){return(0,a.jsxs)("p",{className:h().sourceLine,children:["Snapshot ",(0,a.jsx)("time",{dateTime:d.vN.updated,children:d.vN.updated})," \xb7 LiveBench question set: 25 June 2026 \xb7 ",(0,a.jsx)("a",{href:"/ai-models/methodology#sources",children:"View every source ↗"})," \xb7 ",(0,a.jsx)("a",{href:`/data/ai-models-${d.vN.updated}.json`,download:!0,children:"Download snapshot ↓"})]})}}},e=>{e.O(0,[6235,5497,8430,6514,886,636,6593,8792],()=>e(e.s=13153)),_N_E=e.O()}]);