https://entrim.ai/_next/static/chunks/6e6677c7692ecc30.js

1(globalThis.TURBOPACK||(globalThis.TURBOPACK=[])).push(["object"==typeof document?document.currentScript:void 0,41685,e=>{"use strict";var t=e.i(32846),n=e.i(7284);let a={sm:"text-sm gap-2",md:"text-base gap-2",lg:"text-lg gap-2"};function i({children:e,className:i,size:o="md",variant:s="primary",...r}){return(0,t.jsx)("a",{className:(0,n.default)("custom"!==s&&["font-display inline-flex items-center leading-none font-medium transition",a[o]],"primary"===s&&"hover:text-primary-800 text-gray-900","secondary"===s&&"text-gray-500 hover:text-gray-900",i),...r,children:e})}e.s(["default",()=>i])},4388,e=>{"use strict";var t=e.i(76089),n=e.i(32846);let a=(0,t.forwardRef)((e,t)=>{let{alt:a,color:i="currentColor",size:o="1em",mirrored:s=!1,children:r,...l}=e;return(0,n.jsxs)("svg",{ref:t,xmlns:"http://www.w3.org/2000/svg",width:o,height:o,color:i,fill:"none",viewBox:"0 0 24 24",transform:s?"scale(-1, 1)":void 0,...l,children:[!!a&&(0,n.jsx)("title",{children:a}),r]})});a.displayName="IconBase",e.s(["default",()=>a])},79280,e=>{"use strict";var t=e.i(99247);e.s(["IconPixelChevronRight",()=>t.default])},14650,e=>{"use strict";var t=e.i(32846),n=e.i(7284);let a={primary:"bg-primary-800 text-white shadow-sm transition duration-300 hover:bg-primary-700 focus:ring-primary-500",secondary:"border border-primary-800 bg-white text-primary-800 duration-300 transition hover:bg-primary-800 hover:text-white focus:ring-primary-300",muted:"border border-white/20 bg-white/5 text-white duration-300 transition hover:bg-white/10",gray:"bg-gray-300 text-gray-900 duration-300 transition hover:bg-transparent border border-transparent hover:border-gray-400","outline-gray":"bg-transparent border border-gray-400 text-gray-900 duration-300 transition hover:bg-gray-300 hover:border-transparent",soft:"bg-primary-800/10 text-primary-800 duration-300 transition hover:bg-transparent border border-transparent hover:border-primary-800/30",emerald:"bg-emerald-400 text-primary-1000 duration-300 transition hover:bg-emerald-300"},i={sm:"px-4 h-9 text-base gap-2",md:"px-4 xl:px-6 h-10 text-base gap-2 xl:gap-4",lg:"px-6 h-12 text-base gap-4",xl:"px-6 xl:px-8 h-12 xl:h-16 text-base xl:text-lg gap-4 leading-none",loose:"text-base"};function o({children:e,variant:o="primary",size:s="lg",fullWidth:r=!1,loading:l=!1,className:d,...c}){let m="href"in c&&void 0!==c.href,p="disabled"in c&&c.disabled||l,u=(0,n.default)("inline-flex items-center cursor-pointer justify-center rounded font-display font-medium focus:outline-none [&_svg]:transition-transform [&_svg]:duration-300 hover:[&_svg]:translate-x-1",a[o],i[s],r&&"w-full",!m&&"disabled:cursor-not-allowed disabled:opacity-50",d);if(m){let{href:a,...i}=c;return(0,t.jsx)("a",{href:a,className:(0,n.default)(u,l&&"pointer-events-none opacity-50"),"aria-busy":l,...i,children:l?"Loading...":e})}let{type:h="button",...g}=c;return(0,t.jsx)("button",{type:h,className:u,disabled:p,"aria-busy":l,...g,children:l?"Loading...":e})}e.s(["default",()=>o])},56122,e=>{"use strict";let t={HOME:"/",ABOUT_US:"/about-us",PRICING:"/pricing",AI_MODELS:"/ai-models",DATA_PRIVACY:"/data-privacy",TALK_TO_SALES:"/talk-to-sales",LLM_INFERENCE:"/llm-inference",TERMS_OF_SERVICE:"/terms-of-service",CODING_PLANS:"https://dev.entrim.ai",PRIVACY_POLICY:"/privacy-policy",WAITLIST_SUCCESS:"/waitlist-success",SIGNUP:"/signup",LOGIN:"/login",FORGOT_PASSWORD:"/forgot-password",RESET_PASSWORD:"/reset-password",BLOG_POST:e=>`/blog/${e}`,MODEL_FAMILY:e=>`/ai-models/${e}`,MODEL_DETAILS:(e,t)=>`/ai-models/${e}/${t}`,EXTERNAL:{DISCORD:"https://discord.gg/BW7Zt4jcsj",LINKEDIN:"https://www.linkedin.com/company/entrim/",STATUS:"https://status.entrim.ai",OPENCLAW_PLANS:"https://openclaw.entrim.ai",GPU:"https://gpu.entrim.ai",DOCS:"https://docs.entrim.ai"}};t.HOME,t.ABOUT_US,t.PRICING,t.AI_MODELS,t.DATA_PRIVACY,t.TALK_TO_SALES,t.LLM_INFERENCE,t.TERMS_OF_SERVICE,t.PRIVACY_POLICY,e.s(["ROUTES",0,t])},96120,e=>{"use strict";e.s(["freeCreditAmount",0,10])},61457,61851,e=>{"use strict";var t=e.i(32846),n=e.i(76089),a=e.i(58457),i=e.i(21106);let o="/data/model-performance.json",s=/^\d{4}-\d{2}-\d{2}$/;function r(e){return null===e||"number"==typeof e&&Number.isFinite(e)&&e>=0}function l(e){return!!e&&"object"==typeof e&&"string"==typeof e.date&&s.test(e.date)&&r(e.ttftMs)&&r(e.tokensPerSecond)}function d(e){if(!e||"object"!=typeof e||1!==e.version||"string"!=typeof e.windowStart||!s.test(e.windowStart)||"string"!=typeof e.windowEnd||!s.test(e.windowEnd)||"string"!=typeof e.generatedAt||!e.models||"object"!=typeof e.models)return null;let t={};for(let[n,a]of Object.entries(e.models)){if(!a||"object"!=typeof a)continue;let e=a.days;Array.isArray(e)&&7===e.length&&e.every(l)&&r(a.ttftMs)&&r(a.tokensPerSecond)&&(t[n]={days:e.map(e=>({date:e.date,ttftMs:e.ttftMs,tokensPerSecond:e.tokensPerSecond})),ttftMs:a.ttftMs,tokensPerSecond:a.tokensPerSecond})}return{version:1,generatedAt:e.generatedAt,windowStart:e.windowStart,windowEnd:e.windowEnd,models:t}}

1function c(e,t){let n=Date.parse(`${e.windowEnd}T00:00:00.000Z`)+864e5;return!Number.isFinite(n)||t-n>2592e5}function m(e){let[,t,n]=e.split("-");return`${n}.${t}.`}let p=[1,2,2.5,5,10];function u(e){let t=Math.max(0,...e);if(t<=0)return[0,1,2,3].map(e=>({value:e,label:String(e)}));let n=t/3,a=10**Math.floor(Math.log10(n)),i=(p.find(e=>e*a>=n)??10)*a;return[0,1,2,3].map(e=>{let t=Number((i*e).toPrecision(12));return{value:t,label:t.toLocaleString("en-US")}})}function h(e,t){let n=[],a=[];return e.forEach((i,o)=>{if(null===i){a.length&&n.push(a),a=[];return}a.push({x:e.length>1?o/(e.length-1)*100:50,y:(1-i/t)*100,index:o})}),a.length&&n.push(a),n.map(e=>{let t=e.map((e,t)=>`${0===t?"M":"L"} ${e.x} ${e.y}`).join(" "),n=e[0],a=e[e.length-1];return{line:t,area:`${t} L ${a.x} 100 L ${n.x} 100 Z`,points:e}})}function g(e,t){if(!e||0===t.size)return e;let n={};for(let[a,i]of Object.entries(e.models))t.has(a)||(n[a]=i);return{...e,models:n}}function f(e){let{initialFile:t,seedStale:n,fetched:a,fetchedStale:i,retired:o}=e;return a.file&&!i?{status:"ready",blank:!1,file:g(a.file,o)}:t&&!n?{status:"ready",blank:!1,file:g(t,o)}:a.done?{status:"ready",blank:!1,file:null}:{status:"loading",blank:!0,file:g(t,o)}}e.s(["MODEL_PERFORMANCE_PATH",0,o,"chartSegments",()=>h,"chartTicks",()=>u,"formatPerformanceDay",()=>m,"isModelPerformanceStale",()=>c,"selectPerformanceState",()=>f,"validateModelPerformanceFile",()=>d],61851);let b=(0,n.createContext)({status:"loading",blank:!0,file:null});async function w(){let e=new AbortController,t=setTimeout(()=>e.abort(),5e3);try{let t=await fetch(o,{signal:e.signal,cache:"no-cache"});if(!t.ok)return null;return d(await t.json())}catch{return null}finally{clearTimeout(t)}}function v(e){let t=setInterval(e,36e5);return document.addEventListener("visibilitychange",e),()=>{clearInterval(t),document.removeEventListener("visibilitychange",e)}}function y({initialFile:e=null,children:o}){let{catalog:s}=(0,a.useModelApiCatalog)(),[r,l]=(0,n.useState)({done:!1,file:null}),d=(0,n.useSyncExternalStore)(v,()=>{let t=Date.now(),n=!e||c(e,t),a=!r.file||c(r.file,t);return`${+!!n}${+!!a}`},()=>`${+!e}1`),m="1"===d[0],p="1"===d[1];(0,n.useEffect)(()=>{let e=!1;return w().then(t=>{e||l({done:!0,file:t})}),()=>{e=!0}},[]);let u=r.done&&null!==r.file&&p;(0,n.useEffect)(()=>{if(!u)return;let e=!1;return w().then(t=>{e||l({done:!0,file:t})}),()=>{e=!0}},[u]);let h=(0,n.useMemo)(()=>new Set((s?.models??[]).filter(e=>"retired"===(0,i.catalogModelStatus)(e)).map(e=>e.id)),[s]),g=(0,n.useMemo)(()=>f({initialFile:e,seedStale:m,fetched:r,fetchedStale:p,retired:h}),[r,p,e,m,h]);return(0,t.jsx)(b.Provider,{value:g,children:o})}function k(){return(0,n.useContext)(b)}e.s(["ModelPerformanceProvider",()=>y,"useModelPerformance",()=>k],61457)},26503,91165,e=>{"use strict";e.i(43522);var t=e.i(21106);let n={"deepseek-v4-1-flash":{familyId:"deepseek",badges:[{text:"New"}]},"glm-5-3-flash":{familyId:"glm",badges:[{text:"Turbo",variant:"turbo",tooltip:"Turbo is Entrim's NVFP4 serving mode for GLM 5.3 Flash — the same open weights, quantized only where it is safe, for higher throughput at a lower price."}]},"qwen3-8-27b":{familyId:"qwen"},"deepseek-v4-flash":{familyId:"deepseek"},"gemma-4-26b-a4b":{familyId:"gemma"},"gemma-4-31b":{familyId:"gemma"},"qwen3-embedding-8b":{familyId:"qwen"},"gpt-oss-120b":{familyId:"gpt-oss"},"gpt-oss-20b":{familyId:"gpt-oss"},"glm-5-2":{familyId:"glm"}};function a(e){if(!e)return[];let a=new Map(e.models.map(e=>[e.id,e]));return Object.entries(n).flatMap(([e,n])=>{var i,o;let s,r=a.get(e);return r&&1===r.published&&"retired"!==(0,t.catalogModelStatus)(r)?[(s=r.spec?.quantization??r.quantization,{id:r.id,familyId:n.familyId,logo:r.logo,name:r.name,purpose:r.purpose,price:r.price,context:(r.spec?.context??r.context??"").toUpperCase(),precision:s?.toUpperCase(),highlightCapabilities:r.spec?.highlightCapabilities??[],badges:(i=n.badges,o=(0,t.catalogModelStatus)(r),(0,t.isCatalogModelDeprecated)(o)?[{text:"Deprecated",variant:"tertiary"},...(i??[]).filter(e=>"New"!==e.text)]:function(e,t){if(!t)return e;let n={text:"Beta",variant:"primary"},a=e??[],i=a.findIndex(e=>"Turbo"===e.text);if(i>=0)return[...a.slice(0,i),n,...a.slice(i)];let o=a.findIndex(e=>"New"===e.text);return o>=0?[...a.slice(0,o+1),n,...a.slice(o+1)]:[...a,n]}(i,"beta"===o))})]:[]})}function i(e){return e.modelTableOrder.filter(e=>!(e in n))}e.s(["buildModelLibraryModels",()=>a,"getUncuratedLibraryModelIds",()=>i,"modelFamilies",0,[{id:"qwen",name:"Qwen",logo:"qwen",featured:!0},{id:"gemma",name:"Gemma",logo:"gemma",featured:!0},{id:"deepseek",name:"DeepSeek",logo:"deepseek",featured:!0},{id:"gpt-oss",name:"GPT OSS",logo:"gpt",featured:!1},{id:"kimi",name:"Kimi",logo:"kimi",featured:!1},{id:"glm",name:"GLM",logo:"zai",featured:!1}]],91165);var o=e.i(56122);let s=[{familyId:"deepseek",metaTitle:"DeepSeek API | EU-Hosted, OpenAI-Compatible | Entrim",metaDescription:"Run MIT-licensed DeepSeek models on Entrim's OpenAI-compatible API — EU-only inference, no data to DeepSeek's servers, usage-based pricing, and $10 free credits.",lead:"DeepSeek is a family of MIT-licensed open-weight models with their strongest published results in software engineering — coding agents, tool-calling loops, and long-context document work — at per-token costs far below dense frontier models, because sparse MoE designs activate only a small share of parameters per request.",useCases:[{title:"Coding agents and dev tools",description:"The family’s strongest published benchmarks are software-engineering ones — DeepSeek V4.1 Flash resolves 74.2% of DeepSWE v1.1 issues at maximum reasoning effort.",modelId:"deepseek-v4-1-flash"},{title:"High-volume agent fleets",description:"Sparse MoE activation — 13B of 284B parameters on V4 Flash — keeps per-request cost low across thousands of daily tool calls.",modelId:"deepseek-v4-flash"},{title:"Whole-repo and long-document analysis",description:"A 1M-token context window with sparse attention reads entire repositories and document sets in a single prompt.",modelId:"deepseek-v4-1-flash"},{title:"Document extraction to JSON",description:"Strict function calling and JSON mode return schema-valid output for ticket, PDF, and form pipelines at production volume.",modelId:"deepseek-v4-1-flash"},{title:"RAG and knowledge assistants",description:"Long context fits retrieved passages without aggressive chunking, and thinking modes add reasoning depth only where a query needs it.",modelId:"deepseek-v4-1-flash"},{title:"Batch summarization and classification",description:"Low per-token pricing and deep cache-read discounts suit offline enrichment, tagging, and report generation over large datasets.",modelId:"deepseek-v4-flash"}],exampleApiModelName:"deepseek-ai/DeepSeek-V4.1-Flash",faqs:[{question:"Is the DeepSeek API OpenAI-compatible?",answer:"Yes. Entrim serves DeepSeek models through an OpenAI-compatible endpoint — change the base URL, API key, and model name, and existing chat-completion code, SDKs, and agent frameworks keep working."},{question:"Does my data go to DeepSeek?",answer:["No. Entrim serves DeepSeek's MIT-licensed open-weight models on its own GPU infrastru

1cture in the EU. Prompts and outputs are processed in Entrim's datacenter in Slovenia and never reach DeepSeek's servers.","This is the practical difference from DeepSeek's first-party API, which processes requests on servers in China and offers no EU data residency or DPA. With Entrim you get the same open models under GDPR-compliant, EU-only data handling."]},{question:"Which DeepSeek models can I run on Entrim?",answer:"Entrim serves DeepSeek V4.1 Flash — the 552B-parameter multimodal MoE with a 1M-token context window and a 1–100 reasoning-effort dial — and DeepSeek V4 Flash, the text-only 284B-parameter predecessor with three thinking levels. New DeepSeek releases are evaluated as they ship and added based on customer demand."},{question:"How does the $10 free credit work?",answer:"You can use $10 free credits to test DeepSeek models with your own workload — real prompts, token sizes, and thinking modes — before moving production traffic."},{question:"How does pricing work?",answer:"Pricing is usage-based. You pay for input and output tokens based on the model you use, with discounted cache reads for repeated context."},{question:"How is customer data handled?",answer:"Prompts and outputs are processed for inference and not stored after completion. Customer prompts and outputs are never used to train models. Inference and request processing stay EU-only."},{question:"How do I migrate from my current provider?",answer:"Start by testing one workload. Compare quality, latency, throughput, and request cost against your current provider, then route traffic to Entrim once the workload meets your production target."},{question:"Do you run your own infrastructure?",answer:["Yes. Entrim operates its own datacenter in Slovenia and runs inference on dedicated NVIDIA GPU infrastructure, including B200, H200, and H100 GPUs built for large-scale inference workloads.","The platform runs on an in-house runtime stack designed for high throughput, efficient utilization, and predictable latency. Those efficiency gains are reflected directly in Entrim pricing."]}]},{familyId:"qwen",metaTitle:"Qwen API | Apache-Licensed, EU-Hosted | Entrim",metaDescription:"Run Apache-2.0 Qwen models on Entrim's OpenAI-compatible API — multimodal chat and embedding models, EU-only inference, usage-based pricing, and $10 free credits.",lead:"Qwen is the broadest family of Apache-2.0 open-weight models — natively multimodal chat models, dense and sparse-MoE variants, and top-ranked multilingual embeddings — strongest in tool-heavy agentic coding and multilingual work, and priced low because its MoE models activate only about 3B parameters per request.",useCases:[{title:"Coding agents and dev tools",description:"Qwen 3.8 27B posts 61.7 on SWE-bench Pro and 73.0 on Terminal-Bench 2.1 with thinking enabled — dense, consistent quality for tool-heavy loops.",modelId:"qwen3-8-27b"},{title:"Multimodal document and screenshot pipelines",description:"Vision is native to the family — screenshots, PDFs, diagrams, and video frames are read alongside text, with no separate vision model to route to.",modelId:"qwen3-8-27b"},{title:"Multilingual chat and support",description:"Qwen models are trained across 100+ languages, and thinking control lets one model mix instant replies with deeper handling of escalated cases.",modelId:"qwen3-8-27b"},{title:"RAG and semantic search",description:"Qwen3 Embedding 8B tops multilingual embedding leaderboards — retrieval and generation from one family, one API.",modelId:"qwen3-embedding-8b"},{title:"Document extraction to JSON",description:"Function calling and structured output across the lineup return schema-valid JSON for ticket, PDF, and form pipelines at production volume.",modelId:"qwen3-8-27b"},{title:"Batch summarization and classification",description:"Low per-token pricing and discounted cache reads suit offline enrichment, tagging, and report generation over large datasets.",modelId:"qwen3-8-27b"}],exampleApiModelName:"Qwen/Qwen3.8-27B",faqs:[{question:"Is the Qwen API OpenAI-compatible?",answer:"Yes. Entrim serves Qwen models through an OpenAI-compatible endpoint — change the base URL, API key, and model name, and existing chat-completion code, SDKs, and agent frameworks keep working."},{question:"Does my data go to Alibaba?",answer:["No. Entrim serves Qwen's Apache-2.0 open-weight models on its own GPU infrastru

1cture in the EU. Prompts and outputs are processed in Entrim's datacenter in Slovenia and never reach Alibaba Cloud.","Qwen's first-party API (Alibaba Cloud Model Studio) processes requests in Singapore or mainland China depending on region and tier, with EU residency limited to enterprise arrangements. With Entrim you get the same open models under GDPR-compliant, EU-only data handling by default."]},{question:"Which Qwen models can I run on Entrim?",answer:"Entrim currently serves Qwen 3.8 27B — a dense multimodal model with per-request thinking control — Qwen 3.6 35B-A3B, the sparse MoE with 3B active parameters, and Qwen3 Embedding 8B for multilingual retrieval. New Qwen releases are evaluated as they ship and added based on customer demand."},{question:"How does the $10 free credit work?",answer:"You can use $10 free credits to test Qwen models with your own workload — real prompts, images, token sizes, and thinking modes — before moving production traffic."},{question:"How does pricing work?",answer:"Pricing is usage-based. You pay for input and output tokens based on the model you use, with discounted cache reads for repeated context."},{question:"How is customer data handled?",answer:"Prompts and outputs are processed for inference and not stored after completion. Customer prompts and outputs are never used to train models. Inference and request processing stay EU-only."},{question:"How do I migrate from my current provider?",answer:"Start by testing one workload. Compare quality, latency, throughput, and request cost against your current provider, then route traffic to Entrim once the workload meets your production target."},{question:"Do you run your own infrastructure?",answer:["Yes. Entrim operates its own datacenter in Slovenia and runs inference on dedicated NVIDIA GPU infrastructure, including B200, H200, and H100 GPUs built for large-scale inference workloads.","The platform runs on an in-house runtime stack designed for high throughput, efficient utilization, and predictable latency. Those efficiency gains are reflected directly in Entrim pricing."]}]},{familyId:"gemma",metaTitle:"Gemma API | Google's Open Models, EU-Hosted | Entrim",metaDescription:"Run Apache-2.0 Gemma 4 models on Entrim's OpenAI-compatible API — multimodal, 140+ languages, EU-only inference, usage-based pricing, and $10 free credits.",lead:"Gemma is Google DeepMind's family of Apache-2.0 open-weight models — natively multimodal, pre-trained across 140+ languages, and topping open-model user-preference rankings.",useCases:[{title:"Multilingual chat and support",description:"Out-of-the-box support for 35+ languages and pre-training across 140+ — one deployment serves a global user base, with thinking control for escalated cases.",modelId:"gemma-4-31b"},{title:"Multimodal document and screenshot pipelines",description:"Native vision reads screenshots, PDFs, and diagrams alongside text — grounded answers over 256K tokens of context.",modelId:"gemma-4-31b"},{title:"Coding agents and dev tools",description:"Gemma 4 31B posts 80% on LiveCodeBench and 89% on AIME — dense-model consistency for code review and math-heavy work.",modelId:"gemma-4-31b"},{title:"Human-facing assistants and chat products",description:"At 1452 ELO on LMArena, Gemma 4 31B is the top-ranked open dense model — responses users actually prefer.",modelId:"gemma-4-31b"},{title:"Document extraction to JSON",description:"Function calling and structured output return schema-valid JSON for ticket, PDF, and form pipelines at production volume.",modelId:"gemma-4-31b"},{title:"RAG and knowledge assistants",description:"A 256K context fits retrieved passages and full documents, and configurable thinking adds reasoning depth only where a query needs it.",modelId:"gemma-4-31b"}],exampleApiModelName:"google/gemma-4-31B-it",faqs:[{question:"Is the Gemma API OpenAI-compatible?",answer:"Yes. Entrim serves Gemma models through an OpenAI-compatible endpoint — change the base URL, API key, and model name, and existing chat-completion code, SDKs, and agent frameworks keep working."},{question:"What is the difference between Gemma and Gemini?",answer:["Gemini is Google's proprietary model family, available only through Google-hosted APIs. Gemma is Google DeepMind's open-weight family, released under Apache 2.0 — the weights are public, so independent providers can serve them.","That difference is what makes Entrim possible: you get Google DeepMind model quality with full control over where inference runs and how your data is handled."]},{question:"Does my data go to Google?",answer:["No. Entrim serves Gemma's Apache-2.0 open-weight models on its own GPU infrastru

1cture in the EU. Prompts and outputs are processed in Entrim's datacenter in Slovenia and never reach Google.","Google's hosted Gemma and Gemini endpoints process requests under Google's cloud terms in Google-operated regions. With Entrim, inference stays GDPR-compliant and EU-only by default."]},{question:"Which Gemma models can I run on Entrim?",answer:"Entrim currently serves Gemma 4 31B, the dense flagship with a 256K context window and thinking control. New Gemma releases are evaluated as they ship and added based on customer demand."},{question:"How does the $10 free credit work?",answer:"You can use $10 free credits to test Gemma models with your own workload — real prompts, images, token sizes, and thinking settings — before moving production traffic."},{question:"How does pricing work?",answer:"Pricing is usage-based. You pay for input and output tokens based on the model you use, with discounted cache reads for repeated context."},{question:"How is customer data handled?",answer:"Prompts and outputs are processed for inference and not stored after completion. Customer prompts and outputs are never used to train models. Inference and request processing stay EU-only."},{question:"How do I migrate from my current provider?",answer:"Start by testing one workload. Compare quality, latency, throughput, and request cost against your current provider, then route traffic to Entrim once the workload meets your production target."},{question:"Do you run your own infrastructure?",answer:["Yes. Entrim operates its own datacenter in Slovenia and runs inference on dedicated NVIDIA GPU infrastructure, including B200, H200, and H100 GPUs built for large-scale inference workloads.","The platform runs on an in-house runtime stack designed for high throughput, efficient utilization, and predictable latency. Those efficiency gains are reflected directly in Entrim pricing."]}]}],r=[{modelId:"deepseek-v4-1-flash",familyId:"deepseek",slug:"deepseek-v4-1-flash-api",metaTitle:"DeepSeek V4.1 Flash API | 1M Context, Vision | Entrim",metaDescription:"Run DeepSeek V4.1 Flash on Entrim's OpenAI-compatible API: 1M-token context, image input, 1–100 reasoning effort, EU-only data handling, and $10 free credits.",aboutStats:[{value:"1M",label:"token context window",description:"Sparse attention keeps the global KV cache at 890 bytes per token — about a quarter of DeepSeek V4 Flash."},{value:"90.6",label:"on Terminal-Bench 2.1",description:"Up from 82.7 for DeepSeek V4 Flash at maximum reasoning effort — from MIT-licensed open weights."},{value:"96%",label:"avg. cache hit rate on Entrim",description:"Measured on coding-agent workloads — repeated context bills at the cache-read rate."}],capabilitiesIntro:"DeepSeek V4.1 Flash is strongest where agents reason over a lot of context and now look at screens too: coding loops, tool chains, and document pipelines that need long memory at Flash cost.",capabilities:[{title:"Agentic coding",description:"Holds a plan across long tool-call chains — 74.2 on DeepSWE 1.1, up from 54.4 for V4 Flash, and a 3471 Codeforces rating."},{title:"Native image input",description:"A from-scratch DeepSeek-ViT encoder reads screenshots, diagrams, and scanned pages directly — 95.6 on DocVQA for the base model — with no separate vision model in the loop."},{title:"Reasoning effort from 1 to 100",description:"An integer from 1 to 100 (with low, high, and max presets) sets thinking depth per request, and effort none skips reasoning entirely — pay for depth only where a request needs it."},{title:"Reliable structured output",description:"Strict function calling and JSON mode at every reasoning effort — tool arguments and typed pipelines that parse the first time."}],benchmarks:{items:[{value:"90.6",label:"Terminal-Bench 2.1",description:"Agentic coding and shell tasks run end-to-end in a real terminal."},{value:"74.2",label:"DeepSWE 1.1",description:"Resolves real software-engineering issues across long tool-call chains."},{value:"54.8",label:"AutomationBench",description:"Tool-driven automation across real applications and workflows."}],note:"DeepSeek-reported results at maximum reasoning effort, per the DeepSeek-V4.1-Flash model card (September 2026); independent verification pending."},relatedModelIds:["deepseek-v4-flash","glm-5-3-flash","qwen3-8-27b"]},{modelId:"deepseek-v4-flash",familyId:"deepseek",slug:"deepseek-v4-flash-api",metaTitle:"DeepSeek V4 Flash API | 1M-Token Context | Entrim",metaDescription:"Run DeepSeek V4 Flash on Entrim's OpenAI-compatible API: 1M-token context, three thinking modes, EU-only data handling, and $10 free credits to start.",aboutStats:[{value:"1M",label:"token context window",description:"Sparse attention runs 1M-token prompts on ~27% of the compute of DeepSeek V3.2."},{value:"13B",label:"active of 284B total",description:"MoE rout

1ing keeps per-request cost and latency near small-model levels."},{value:"97%",label:"avg. cache hit rate on Entrim",description:"Measured on coding-agent workloads — repeated context bills at the cache-read rate."}],capabilitiesIntro:"DeepSeek V4 Flash is strongest where agents reason over a lot of context: coding loops, tool chains, and document pipelines that need long memory without dense-model cost.",capabilities:[{title:"Agentic coding",description:"Holds a plan across long tool-call chains — the 0731 re-post-train pushes Terminal-Bench 2.1 to 82.7, up from 61.8 on the preview build."},{title:"Three thinking levels",description:"Non-think, think-high, and think-max — the 0731 re-post-train sharpens the reasoning modes hardest (DeepSWE 7.3 → 54.4), so you pay for depth only where a request needs it."},{title:"1M-token context",description:"Hybrid sparse attention reads 1M tokens with ~10% of the KV cache of V3.2 — whole repos and document sets in one prompt."},{title:"Reliable structured output",description:"Strict function calling and JSON mode in every thinking level — tool arguments and typed pipelines that parse the first time."}],benchmarks:{items:[{value:"82.7",label:"Terminal-Bench 2.1",description:"Agentic coding and shell tasks run end-to-end in a real terminal."},{value:"54.4",label:"DeepSWE",description:"Resolves real software-engineering issues across long tool-call chains."},{value:"70.3",label:"Toolathlon (verified)",description:"Sustains long multi-tool workflows without losing the plan."}],note:"Self-reported by DeepSeek for the V4 Flash 0731 build (July 2026); independent verification pending."},relatedModelIds:["deepseek-v4-1-flash","qwen3-8-27b","gemma-4-31b"]},{modelId:"qwen3-6-35b-a3b",familyId:"qwen",slug:"qwen3-6-35b-a3b-api",metaTitle:"Qwen 3.6 35B-A3B API | Multimodal MoE, 3B Active | Entrim",metaDescription:"Run Qwen 3.6 35B-A3B on Entrim's OpenAI-compatible API — multimodal MoE with 3B active parameters, 262K context, EU-only data handling, and $10 free credits.",aboutStats:[{value:"3B",label:"active of 35B total",description:"A 256-expert MoE routes 8 experts per token — near-dense quality at small-model compute."},{value:"3",label:"input modalities",description:"Text, images, and video are read natively — no separate vision model to route to."},{value:"97%",label:"avg. cache hit rate on Entrim",description:"Measured on coding-agent workloads — repeated context bills at the cache-read rate."}],capabilitiesIntro:"Qwen 3.6 35B-A3B is strongest where agents call tools constantly and the context includes more than text — coding loops, visual tickets, and multilingual pipelines at fleet scale.",capabilities:[{title:"Agentic coding",description:"73.4% on SWE-bench Verified and 51.5% on Terminal-Bench 2.0 with thinking enabled — plans, edits, and verifies real repositories."},{title:"Native multimodal input",description:"81.7 on MMMU and 75.3 on MMMU-Pro — screenshots, diagrams, and video frames are first-class inputs, not an add-on."},{title:"Thinking preservation",description:"Reasoning traces persist across conversation turns, so multi-step agent loops keep earlier decisions without re-deriving them."},{title:"Competition-grade math",description:"92.7 on AIME 2026 and 83.6 on HMMT with thinking enabled — reasoning depth on demand, off when speed matters."}],benchmarks:{items:[{value:"73.4%",label:"SWE-bench Verified",description:"Resolves real GitHub issues end-to-end in agentic coding runs."},{value:"80.4",label:"LiveCodeBench v6",description:"Competitive coding problems unseen at training time."},{value:"81.7",label:"MMMU",description:"University-level reasoning over images, charts, and diagrams."}],note:"Published results for Qwen 3.6 35B-A3B with thinking enabled, per the Qwen model card and Artificial Analysis."},relatedModelIds:["qwen3-8-27b","gemma-4-26b-a4b","deepseek-v4-flash"]},{modelId:"qwen3-8-27b",familyId:"qwen",slug:"qwen3-8-27b-api",metaTitle:"Qwen 3.8 27B API | Dense Agentic Coding | Entrim",metaDescription:"Run Qwen 3.8 27B on Entrim's OpenAI-compatible API — a dense multimodal coder with 262K context, thinking control, EU-only data handling, and $10 free credits.",aboutStats:[{value:"27B",label:"dense parameters",description:"Every parameter is active on every token — 

1consistent quality with no routing variance."},{value:"84.3",label:"on OSWorld-Verified",description:"Qwen-reported autonomous desktop computer use — screenshots in, actions out."},{value:"97%",label:"avg. cache hit rate on Entrim",description:"Measured on coding-agent workloads — repeated context bills at the cache-read rate."}],capabilitiesIntro:"Qwen 3.8 27B is strongest on agentic coding and long-horizon computer-use work — repository patches, visual front-end tickets, and multi-turn agents that keep earlier reasoning.",capabilities:[{title:"Agentic coding",description:"Qwen reports 61.7 on SWE-bench Pro and 73.0 on Terminal-Bench 2.1 with thinking enabled — plans, edits, and verifies real repositories."},{title:"Long-horizon coding endurance",description:"Qwen reports DeepSWE 1.1 jumping 13.3 → 42.2 over the previous generation — multi-step fixes hold to the end of the run."},{title:"Thinking control",description:"On by default and switchable per request — reasoning_effort (xhigh, medium, low) and preserve_thinking tune depth and carry traces across turns."},{title:"Native multimodal input",description:"Screenshots, diagrams, and hour-scale video are first-class inputs — front-end bugs and visual tickets stay in one model."}],benchmarks:{items:[{value:"61.7",label:"SWE-bench Pro",description:"Harder, contamination-resistant software engineering tasks."},{value:"73.0",label:"Terminal-Bench 2.1",description:"Autonomous work in a real terminal — builds, debugging, ops tasks."},{value:"42.2",label:"DeepSWE 1.1",description:"Long-horizon agentic coding with environment feedback."}],note:"Qwen-reported results with thinking enabled — SWE-bench Pro via the Claude Code harness, Terminal-Bench 2.1 via Terminus, temp=1.0, top_p=0.95 — per the Qwen model card. Independent labs have not reproduced these scores."},relatedModelIds:["glm-5-3-flash","deepseek-v4-flash","gemma-4-31b"]},{modelId:"gemma-4-26b-a4b",familyId:"gemma",slug:"gemma-4-26b-a4b-api",metaTitle:"Gemma 4 26B A4B API | 4B-Active Multimodal MoE | Entrim",metaDescription:"Run Gemma 4 26B A4B on Entrim's OpenAI-compatible API — Google's 4B-active multimodal MoE with 256K context, EU-only data handling, and $10 free credits.",aboutStats:[{value:"3.8B",label:"active of 25.2B total",description:"128 fine-grained experts route per token — near-31B quality at small-model compute."},{value:"256K",label:"token context window",description:"Holds long documents, retrieved passages, and multi-turn sessions in one prompt."},{value:"140+",label:"languages in pre-training",description:"35+ supported out of the box — one deployment serves a global user base."}],capabilitiesIntro:"Gemma 4 26B A4B is strongest where volume meets variety — multilingual assistants, visual document pipelines, and high-throughput extraction that need flagship-adjacent quality without flagship cost.",capabilities:[{title:"Near-flagship quality",description:"82.6% on MMLU-Pro and an Arena ELO within a dozen points of Gemma 4 31B — from a fraction of the compute per token."},{title:"Native vision",description:"73.8 on MMMU-Pro — screenshots, PDFs, and diagrams are read alongside text, no separate vision model to route to."},{title:"Configurable thinking",description:"Step-by-step reasoning on demand — 88.3 on AIME 2026 with thinking enabled, fast direct replies with it off."},{title:"Reliable structured output",description:"Strict function calling and JSON mode — tool arguments and extraction pipelines that parse the first time."}],benchmarks:{items:[{value:"82.6%",label:"MMLU-Pro",description:"Graduate-level knowledge and reasoning across 14 domains."},{value:"73.8",label:"MMMU-Pro",description:"University-level reasoning over images, charts, and diagrams."},{value:"88.3",label:"AIME 2026",description:"Competition mathematics measuring multi-step reasoning."}],note:"Published results for Gemma 4 26B A4B with thinking enabled, per the Gemma 4 technical report and Artificial Analysis."},relatedModelIds:["gemma-4-31b","qwen3-8-27b","gpt-oss-20b"]},{modelId:"gemma-4-31b",familyId:"gemma",slug:"gemma-4-31b-api",metaTitle:"Gemma 4 31B API | Top-Ranked Open Dense Model | Entrim",metaDescription:"Run Gemma 4 31B on Entrim's OpenAI-compatible API — Google's dense multimodal flagship with 256K context, EU-only data handling, and $10 free credits.",aboutStats:[{value:"1452",label:"LMArena ELO",description:"The top-ranked open dense model in human preference evaluations."},{value:"256K",label:"token context window",description:"Holds long documents, retrieved passages, and multi-turn sessions in one prompt."},{value:"140+",label:"languages in pre-training",description:"35+ supported out of the box — one deployment serves a global user base."}],capabilitiesIntro:"Gemma 4 31B is strongest where answer quality is judged by people — assistants, document analysis, and multilingual products that need flagship reasoning with native vision.",capabilities:[{title:"Flagship reasoning",description:"85.2% on MMLU-Pro and 89.2 on AIME 2026 with thinking enabled — reasoning depth on demand, direct replies with it off."},{title:"Native vision",description:"76.9 on MMMU-Pro — screenshots, PDFs, and diagrams are read alongside text, no separate vision model to route to."},{title:"Strong coding",description:"80.0 on LiveCodeBench v6 — reviews, patches, and test generation with 256K of repository context in scope."},{title:"Reliable structured output",description:"Strict function calling and JSON mode — tool arguments and extraction pipelines that parse the first time."}],benchmarks:{items:[{value:"85.2%",label:"MMLU-Pro",description:"Graduate-level knowledge and reasoning across 14 domains."},{value:"89.2",label:"AIME 2026",description:"Competition mathematics measuring multi-step reasoning."}

1,{value:"80.0",label:"LiveCodeBench v6",description:"Competitive coding problems unseen at training time."}],note:"Published results for Gemma 4 31B with thinking enabled, per the Gemma 4 technical report and Artificial Analysis."},relatedModelIds:["gemma-4-26b-a4b","qwen3-8-27b","deepseek-v4-flash"]},{modelId:"qwen3-embedding-8b",familyId:"qwen",slug:"qwen3-embedding-8b-api",metaTitle:"Qwen3 Embedding 8B API | #1 MTEB Multilingual | Entrim",metaDescription:"Run Qwen3 Embedding 8B on Entrim's OpenAI-compatible API — top-ranked multilingual embeddings for search and RAG, EU-only data handling, and $10 free credits.",aboutStats:[{value:"#1",label:"MTEB multilingual at release",description:"Score 70.58 on the leaderboard spanning retrieval, classification, and clustering."},{value:"4096",label:"embedding dimensions",description:"Matryoshka support scales vectors down to 32 dims to cut storage and search cost."},{value:"100+",label:"languages supported",description:"One embedding space across natural and programming languages."}],capabilitiesIntro:"Qwen3 Embedding 8B is the retrieval side of a production RAG stack — search, deduplication, and classification that stay accurate across languages and code.",capabilities:[{title:"Multilingual retrieval",description:"Cross-language matching in 100+ languages — query in German, retrieve English documents, and vice versa."},{title:"Code search",description:"81.22 on MTEB-Code — natural-language queries over repositories, snippets, and documentation."},{title:"Instruction-aware vectors",description:"A one-line task instruction tunes embeddings for retrieval, classification, or clustering — no fine-tuning required."},{title:"Matryoshka dimensions",description:"Truncate 4096-dim vectors down to 32 with graceful quality falloff — match vectors to your store budget."}],benchmarks:{items:[{value:"70.58",label:"MTEB Multilingual",description:"No. 1 on the multilingual embedding leaderboard at release."},{value:"81.22",label:"MTEB-Code",description:"Code retrieval from natural-language queries over source and docs."},{value:"32K",label:"input tokens per request",description:"Long passages embed whole — less chunking, more context per vector."}],note:"Published results for Qwen3 Embedding 8B, per the Qwen3 Embedding technical report and the MTEB leaderboard."},relatedModelIds:["qwen3-8-27b","deepseek-v4-flash","gemma-4-26b-a4b"]},{modelId:"gpt-oss-120b",familyId:"gpt-oss",slug:"gpt-oss-120b-api",metaTitle:"GPT OSS 120B API | OpenAI Open-Weight Reasoning | Entrim",metaDescription:"Run GPT OSS 120B on Entrim's OpenAI-compatible API — OpenAI's open-weight reasoning MoE with adjustable effort, EU-only data handling, and $10 free credits.",aboutStats:[{value:"5.1B",label:"active of 117B total",description:"MoE routing makes it one of the fastest open models served at scale."},{value:"3",label:"reasoning effort levels",description:"Low, medium, and high — AIME 2024 jumps from 56.3% to 95.8% as effort rises."},{value:"131K",label:"token context window",description:"Fits long documents and full tool-call transcripts in a single request."}],capabilitiesIntro:"GPT OSS 120B is strongest where request difficulty varies — reasoning APIs, analysis endpoints, and tool-calling backends that need depth on some calls and speed on the rest.",capabilities:[{title:"Configurable reasoning effort",description:"One dial per request — AIME 2024 climbs 56.3 → 80.4 → 95.8 across low, medium, and high effort."},{title:"Graduate-level science",description:"80.1% on GPQA Diamond at high effort — analysis that holds up in technical and scientific domains."},{title:"Fast tool calling",description:"Top-3 output speed among open models on Artificial Analysis, with Harmony-trained function calling."},{title:"Reliable structured output",description:"Strict JSON schemas and tool arguments that parse the first time — built for API backends."}],benchmarks:{items:[{value:"95.8%",label:"AIME 2024",description:"Competition mathematics measuring multi-step reasoning."},{value:"80.1%",label:"GPQA Diamond",description:"Graduate-level scientific reasoning and analysis."}

1,{value:"62.4%",label:"SWE-bench Verified",description:"Real GitHub issues resolved end-to-end in agentic runs."}],note:"Published results for GPT OSS 120B at high reasoning effort, per the OpenAI model card and Artificial Analysis."},relatedModelIds:["gpt-oss-20b","deepseek-v4-flash","qwen3-8-27b"]},{modelId:"gpt-oss-20b",familyId:"gpt-oss",slug:"gpt-oss-20b-api",metaTitle:"GPT OSS 20B API | Low-Latency Open Reasoning | Entrim",metaDescription:"Run GPT OSS 20B on Entrim's OpenAI-compatible API — OpenAI's compact reasoning MoE with adjustable effort, EU-only data handling, and $10 free credits.",aboutStats:[{value:"3.6B",label:"active of 21B total",description:"MoE routing delivers real reasoning at the smallest serving footprint."},{value:"3",label:"reasoning effort levels",description:"Low, medium, and high — accuracy scales smoothly as effort rises."},{value:"131K",label:"token context window",description:"Long documents and full tool-call transcripts fit in a single request."}],capabilitiesIntro:"GPT OSS 20B is strongest where volume and latency dominate — classification, routing, extraction, and assistant backends that need reasoning quality without reasoning prices.",capabilities:[{title:"Configurable reasoning effort",description:"A one-line system-message setting trades latency for depth — smooth accuracy scaling across low, medium, and high."},{title:"Competition-grade math",description:"98.7% on AIME 2025 with tools at high effort — from a model that fits on 16GB devices."},{title:"Fast structured output",description:"Harmony-trained function calling and strict JSON — tool arguments that parse the first time, at top-15 open-model speed."},{title:"Escalation-ready",description:"Shares the Harmony format with GPT OSS 120B — route hard requests up without changing your integration."}],benchmarks:{items:[{value:"98.7%",label:"AIME 2025 (with tools)",description:"Competition mathematics measuring multi-step reasoning."},{value:"71.5%",label:"GPQA",description:"Graduate-level scientific reasoning and analysis."},{value:"165",label:"tokens/s (Artificial Analysis)",description:"Top-15 fastest of the 130 open models measured."}],note:"Published results for GPT OSS 20B at high reasoning effort, per the OpenAI model card and Artificial Analysis."},relatedModelIds:["gpt-oss-120b","deepseek-v4-flash","gemma-4-26b-a4b"]},{modelId:"glm-5-3-flash",familyId:"glm",slug:"glm-5-3-flash-api",metaTitle:"GLM 5.3 Flash API | Multimodal 1M Context | Entrim",metaDescription:"Run GLM 5.3 Flash on Entrim's OpenAI-compatible API — Z.ai's MIT-licensed multimodal MoE with 1M-token context, EU-only data handling, and $10 free credits.",aboutStats:[{value:"18B",label:"active of 320B parameters",description:"Sparse routing keeps per-token compute near a small model while the full expert pool stays available."},{value:"84.3",label:"on Terminal-Bench 2.1",description:"Within 0.7 points of Claude Opus 4.8 on end-to-e

1nd terminal tasks — from MIT-licensed open weights."},{value:"97%",label:"avg. cache hit rate on Entrim",description:"Measured on coding-agent workloads — repeated context bills at the cache-read rate."}],capabilitiesIntro:"GLM 5.3 Flash is strongest where agents run long, look at screens, and have to stay cheap — coding and automation loops that mix code, screenshots, and recordings in one context.",capabilities:[{title:"Native image and video input",description:"Trained on a 30T-token multimodal corpus, it reads screenshots, diagrams, and screen recordings directly — no separate vision model in the loop."},{title:"1M tokens without the memory bill",description:"Linear KDA layers plus NoPE sparse MLA cut attention compute by about 3× and the KV cache by 4.4× versus the flagship GLM 5.3 — whole-repo prompts run at full context."},{title:"Selectable reasoning effort",description:"Low, high, and max efforts set thinking depth per request — deep reasoning for the hard steps, shorter traces where a request does not need them."},{title:"Strict tool calling and JSON output",description:"Function calling and structured JSON output in every mode, with 48.8 on AutomationBench — nearly double GLM 5.2 — for multi-step tool use."}],benchmarks:{items:[{value:"84.3",label:"Terminal-Bench 2.1",description:"Agentic coding and shell tasks run end-to-end in a real terminal."},{value:"63.4",label:"DeepSWE 1.1",description:"Multi-step software engineering fixes — up from 46.2 for GLM 5.2."},{value:"48.8",label:"AutomationBench",description:"Tool-driven automation across real applications and workflows."}],note:"Z.ai-reported results at maximum reasoning effort, per the GLM 5.3 Flash launch notes and model card."},serving:{title:"What Turbo means",intro:"Turbo is Entrim's NVFP4 serving mode for GLM 5.3 Flash — the same open weights, quantized only where it is safe, for higher throughput at a lower price.",mechanism:["Only the routed-expert weights — about 97% of the model’s parameters — are stored in NVFP4. Attention layers, routers, embeddings, the vision tower, and all activations stay in higher precision, so 4-bit noise never touches the layers most sensitive to it.","That cuts the weight footprint by about 70%, which is what makes the 1M-token context practical to serve at Flash pricing."],comparison:{label:"BF16 vs NVFP4, measured by NVIDIA on GLM 5.3 Flash",baselineLabel:"BF16",servedLabel:"NVFP4",rows:[{benchmark:"GPQA Diamond",baseline:"92.17",served:"92.11",delta:"−0.06"},{benchmark:"SciCode",baseline:"56.21",served:"57.69",delta:"+1.48"},{benchmark:"MMMU Pro",baseline:"76.88",served:"76.30",delta:"−0.58"},{benchmark:"AA-LCR (long context)",baseline:"71.00",served:"71.06",delta:"+0.06"},{benchmark:"IFBench",baseline:"61.30",served:"60.54",delta:"−0.76"},{benchmark:"Terminal Bench 2.1",baseline:"82.58",served:"83.15",delta:"+0.57"}],note:"Measured by NVIDIA on its GLM 5.3 Flash NVFP4 checkpoint at temperature 1.0, top-p 0.95. Entrim’s own evals will replace this table when they land."},caveat:"The table above is NVIDIA's measurement, not Entrim's — validate outputs against your own evals before moving production traffic."},relatedModelIds:["deepseek-v4-flash","glm-5-2","qwen3-8-27b"]},{modelId:"glm-5-2",familyId:"glm",slug:"glm-5-2-api",metaTitle:"GLM 5.2 API | Frontier-Grade Coding Agents | Entrim",metaDescription:"Run GLM 5.2 on Entrim's OpenAI-compatible API — Z.ai's MIT-licensed 744B MoE with 1M-token context, frontier-grade coding, EU-only data handling, and $10 free credits.",aboutStats:[{value:"62.1",label:"on SWE-bench Pro",description:"Ahead of GPT-5.5 (58.6) — the strongest open-weight result on the hardest coding set."},{value:"1M",label:"token context window",description:"5× the GLM 5.1 window — an entire codebase or document set in one prompt."},{value:"#3",label:"on the Intelligence Index",description:"Scores 53 on Artificial Analysis — ranked above every other model in the Entrim catalog."}],capabilitiesIntro:"GLM 5.2 is strongest where engineering runs are long and the bar is frontier-level — project-scale refactors, tool-driven automation, and client-side work that has to satisfy real users.",capabilities:[{title:"Frontier-parity engineering",description:"62.1 on SWE-bench Pro — past GPT-5.5 on the contamination-resistant set, from MIT-licensed open weights."},{title:"A generational agentic jump",description:"Terminal-Bench 2.1 climbs 62.0 → 81.0 over GLM 5.1 — end-to-e

1nd terminal work that used to need a closed model."},{title:"Design-grade front-end output",description:"Tops Design Arena rankings ahead of closed frontier models — client-side and mobile engineering is this release’s called-out step change."},{title:"Selectable reasoning effort",description:"High and max efforts set thinking depth per request — deep reasoning for the hard steps, faster replies where a request doesn’t need it."}],benchmarks:{items:[{value:"62.1",label:"SWE-bench Pro",description:"Contamination-resistant software engineering — ahead of GPT-5.5 at 58.6."},{value:"81.0",label:"Terminal-Bench 2.1",description:"Agentic coding and shell tasks run end-to-end in a real terminal."},{value:"77.0",label:"MCP-Atlas",description:"Tool use through the Model Context Protocol across multi-step workflows."}],note:"Published results for GLM 5.2 at maximum reasoning effort, per the Z.ai release and Artificial Analysis."},relatedModelIds:["glm-5-3-flash","deepseek-v4-flash","qwen3-8-27b"]}];function l(e){return s.find(t=>t.familyId===e)?o.ROUTES.MODEL_FAMILY(e):void 0}function d(e,t){return r.find(n=>n.familyId===e&&n.slug===t)}function c(e){let t=r.find(t=>t.modelId===e);return t?o.ROUTES.MODEL_DETAILS(t.familyId,t.slug):void 0}e.s(["getFamilyPagePath",()=>l,"getModelDetailPath",()=>c,"getModelPageByPath",()=>d],26503)},37240,2341,e=>{"use strict";e.s(["PAGE_SECTION_IDS",0,{MODELS:"section-models",PRICING_TABLE:"section-pricing-table",MODEL_DETAIL_SAVINGS:"section-model-detail-savings",MODEL_DETAIL_SERVING:"section-what-turbo-means",CALL_TO_ACTION:"section-call-to-action-form",CALL_TO_ACTION_DRAWER:"section-call-to-action-form-2"}],37240);let t={behavior:"smooth",block:"start"};function n(e,n){let a=document.getElementById(e);return!!a&&(a.scrollIntoView({...t,...n}),!0)}e.s(["scrollToSection",()=>n],2341)}]);