[{"data":1,"prerenderedAt":648},["ShallowReactive",2],{"blog-post-mimo-v2-6-pro-agents-setup-cost":3,"related-posts-mimo-v2-6-pro-agents-setup-cost":623},{"id":4,"title":5,"author":6,"body":10,"category":597,"date":598,"description":599,"extension":600,"featured":601,"hideToc":601,"image":602,"imageAlt":603,"imageHeight":604,"imageWidth":605,"itemList":606,"lang":606,"meta":607,"navigation":608,"noindex":601,"path":609,"readingTime":610,"redirected":601,"relatedSlugs":606,"seo":611,"seoTitle":5,"stem":612,"tags":613,"updatedDate":606,"__hash__":622},"blog/blog/mimo-v2-6-pro-agents-setup-cost.md","MiMo-V2.6-Pro for AI Agents: Setup, Cost vs GLM and MiniMax",{"name":7,"role":8,"avatar":9},"Shabnam Katoch","Growth Head","/img/avatars/shabnam-profile.jpeg",{"type":11,"value":12,"toc":576},"minimark",[13,17,20,23,26,31,34,37,40,43,50,54,57,60,63,66,69,72,76,79,196,202,205,208,259,262,271,277,281,284,299,309,315,320,323,398,405,409,412,418,422,425,436,439,449,453,456,459,462,465,469,475,481,487,495,499,502,505,508,516,520,523,526,530,533,537,551,555,558,562,565,569,572],[14,15,16],"p",{},"The new top open-weight model costs less per task than models half as smart. It's also slow, wordy, and comes with a footnote you should read before you plug it in.",[14,18,19],{},"Your agent runs on GLM-5.3. It's good. The invoice is less good.",[14,21,22],{},"Then MiMo-V2.6-Pro lands on 21 September, scores one point higher on the main independent leaderboard, and costs about a third as much per input token. Your first thought is swap it in tonight.",[14,24,25],{},"Hold that thought for ten minutes. We run agents on dozens of models across 28+ providers, so we went through MiMo-V2.6-Pro the way you'd actually use it: in an agent loop, with tools, on a budget.",[27,28,30],"h2",{"id":29},"what-is-mimo-v26-pro","What is MiMo-V2.6-Pro?",[14,32,33],{},"MiMo-V2.6-Pro is Xiaomi's open-weight flagship model, released 21 September 2026 under the MIT licence. It's a sparse mixture-of-experts model with 1.02 trillion total parameters and 42 billion active per token, a 1M-token context window, and text, image, video and audio input.",[14,35,36],{},"Xiaomi's pitch is reinforcement learning at scale. The model card describes one mixed RL run across coding, general agents, visual tasks and cybersecurity, instead of separate runs per domain.",[14,38,39],{},"It ships alongside three siblings. MiMo-V2.6-Flash is the cheaper 309B model with 15B active. Pro-UltraSpeed is API-only and much faster for 10x the price. And there's a small 9B distilled checkpoint for research.",[14,41,42],{},"MiMo-V2.6-Pro is the highest-scoring open-weight model on Artificial Analysis right now. It is not the highest-scoring model, full stop.",[14,44,45],{},[46,47],"img",{"alt":48,"src":49},"The MiMo-V2.6 family, four models in one release: Pro (1.02T total, 42B active, $0.435/$0.87), Flash (309B, 15B active, $0.14/$0.28), Pro UltraSpeed (API only, 10x the price) and Distill 9B for research. All open weights except UltraSpeed.","/img/blog/mimo-v2-6-pro-agents-setup-cost-four-models.jpg",[27,51,53],{"id":52},"mimo-v26-pro-benchmark-results-the-honest-read","MiMo V2.6 Pro benchmark results: the honest read",[14,55,56],{},"There are two sets of numbers. Give the independent one more weight.",[14,58,59],{},"Artificial Analysis scores MiMo-V2.6-Pro at 46 on its Intelligence Index, the top open-weight score it tracks. GLM-5.3 (Max) sits at 45, GLM-5.3 Flash at 42, DeepSeek V4.1 Flash at 39, MiMo-V2.6-Flash at 38 and MiniMax M3 at 29.",[14,61,62],{},"So against GLM-5.3, the lead is one point. Call it a tie. Against MiniMax M3, it's a different league.",[14,64,65],{},"Xiaomi's own model card is more interesting for agent builders. On Toolathlon-Verified, a tool-use benchmark, Pro scores 76.9. On Terminal Bench 2.1 it scores 89.9, level with Claude Opus 5's 89.1. But on the harder Terminal Bench 4.0, it drops to 34.9 against Opus 5's 49.0.",[14,67,68],{},"Here's the weird part. The jump from V2.5-Pro is enormous on agent tasks. DeepSWE goes from 19.0 to 71.9. AutomationBench from 16.0 to 53.1. Same model size. The RL recipe did that.",[14,70,71],{},"Two caveats. Those rows are vendor-reported, run in Xiaomi's own test setups. And Artificial Analysis flags the model as notably slow, at about 43 output tokens per second when we checked, and verbose: it generated 140M tokens to complete the index.",[27,73,75],{"id":74},"mimo-v26-pro-pricing-vs-glm-and-minimax","MiMo V2.6 Pro pricing vs GLM and MiniMax",[14,77,78],{},"Per token, MiMo-V2.6-Pro is cheap. Per finished task, it's even better.",[80,81,82,104],"table",{},[83,84,85],"thead",{},[86,87,88,92,95,98,101],"tr",{},[89,90,91],"th",{},"Model",[89,93,94],{},"Input / 1M",[89,96,97],{},"Output / 1M",[89,99,100],{},"AA index",[89,102,103],{},"AA cost per index task",[105,106,107,128,145,162,179],"tbody",{},[86,108,109,116,119,122,125],{},[110,111,112],"td",{},[113,114,115],"strong",{},"MiMo-V2.6-Pro",[110,117,118],{},"$0.435",[110,120,121],{},"$0.87",[110,123,124],{},"46",[110,126,127],{},"$0.13",[86,129,130,133,136,139,142],{},[110,131,132],{},"MiMo-V2.6-Flash",[110,134,135],{},"$0.14",[110,137,138],{},"$0.28",[110,140,141],{},"38",[110,143,144],{},"$0.06",[86,146,147,150,153,156,159],{},[110,148,149],{},"GLM-5.3 (Max)",[110,151,152],{},"$1.40",[110,154,155],{},"$4.40",[110,157,158],{},"45",[110,160,161],{},"$2.01",[86,163,164,167,170,173,176],{},[110,165,166],{},"GLM-5.3 Flash",[110,168,169],{},"$0.15",[110,171,172],{},"$0.50",[110,174,175],{},"42",[110,177,178],{},"$0.25",[86,180,181,184,187,190,193],{},[110,182,183],{},"MiniMax M3",[110,185,186],{},"$0.30",[110,188,189],{},"$1.20",[110,191,192],{},"29",[110,194,195],{},"$0.51",[14,197,198],{},[199,200,201],"em",{},"Prices are list prices as shown on OpenRouter and Artificial Analysis on 5 October 2026. Cost per task is Artificial Analysis's measured average across its evaluations.",[14,203,204],{},"Stay with me here, because the last column changes the story. GLM-5.3 costs about 15x more per finished task than MiMo-V2.6-Pro for a one-point lower score. And MiMo-V2.6-Pro beats GLM-5.3 Flash on both score and cost per task, even though Flash looks cheaper per token.",[14,206,207],{},"Now in agent terms. Using the same method as our other model pages, a typical agent task reads about 20,000 tokens and writes about 2,000. At 10 tasks a day, that's roughly 300 a month:",[80,209,210,219],{},[83,211,212],{},[86,213,214,216],{},[89,215,91],{},[89,217,218],{},"Light use, 300 tasks a month",[105,220,221,228,235,242,251],{},[86,222,223,225],{},[110,224,132],{},[110,226,227],{},"~$1.00",[86,229,230,232],{},[110,231,166],{},[110,233,234],{},"~$1.20",[86,236,237,239],{},[110,238,183],{},[110,240,241],{},"~$2.50",[86,243,244,248],{},[110,245,246],{},[113,247,115],{},[110,249,250],{},"~$3.10",[86,252,253,256],{},[110,254,255],{},"GLM-5.3",[110,257,258],{},"~$11.00",[14,260,261],{},"Those are upper bounds without caching. MiMo's cached input price is $0.0036 per million, so an agent with a long, stable system prompt pays close to nothing for the repeated part. If 15,000 of those 20,000 input tokens are the same instructions every run, MiMo-V2.6-Pro's light-use bill drops from about $3.10 to about $1.20. The catch is verbosity. If your agent thinks out loud for 8,000 tokens instead of 2,000, the output line grows, so measure on your own tasks.",[14,263,264,265,270],{},"We broke down the same trade-off for the previous generation in our ",[266,267,269],"a",{"href":268},"/blog/minimax-m3-vs-glm-vs-claude-cost-breakdown","MiniMax M3 vs GLM vs Claude cost breakdown",", if you want the longer history.",[14,272,273],{},[46,274],{"alt":275,"src":276},"Cheaper per task, higher score: Artificial Analysis Intelligence Index against cost per task, October 2026, with MiMo Pro at 46 in the top left, GLM-5.3 at 45 far to the right, GLM-5.3 Flash 42, MiMo Flash 38 and MiniMax M3 29.","/img/blog/mimo-v2-6-pro-agents-setup-cost-cost-vs-score.jpg",[27,278,280],{"id":279},"how-to-set-up-mimo-v26-pro-for-an-agent","How to set up MiMo-V2.6-Pro for an agent",[14,282,283],{},"You have three realistic routes. Self-hosting Pro isn't one of them for most people.",[14,285,286,289,290,294,295,298],{},[113,287,288],{},"Through Xiaomi's API."," Xiaomi serves the model on its MiMo API platform under the model ID ",[291,292,293],"code",{},"mimo-v2.6-pro",", with ",[291,296,297],{},"mimo-v2.6-flash"," for the cheaper variant. Xiaomi recommends temperature 1.0 and top_p 0.95.",[14,300,301,304,305,308],{},[113,302,303],{},"Through OpenRouter."," One key, model ID ",[291,306,307],{},"xiaomi/mimo-v2.6-pro",". OpenRouter lists four providers serving it, including Xiaomi itself, and marks it as supporting tools.",[14,310,311,314],{},[113,312,313],{},"Self-hosting."," The Pro checkpoint is about 573 GB, so you need a full 8-GPU node of H200-class cards before you've stored any context. Flash is about 178 GB, which is one server.",[316,317,319],"h3",{"id":318},"the-tool-calling-gotcha","The tool calling gotcha",[14,321,322],{},"This is where most people get it wrong. If you serve MiMo with vLLM yourself, you must pass the MiMo-specific parsers:",[324,325,330],"pre",{"className":326,"code":327,"language":328,"meta":329,"style":329},"language-bash shiki shiki-themes github-light","vllm serve XiaomiMiMo/MiMo-V2.6-Pro-RL \\\n  --tensor-parallel-size 8 \\\n  --trust-remote-code \\\n  --reasoning-parser mimo \\\n  --tool-call-parser mimo \\\n  --enable-auto-tool-choice\n","bash","",[291,331,332,352,363,371,382,392],{"__ignoreMap":329},[333,334,337,341,345,348],"span",{"class":335,"line":336},"line",1,[333,338,340],{"class":339},"s7eDp","vllm",[333,342,344],{"class":343},"sYBdl"," serve",[333,346,347],{"class":343}," XiaomiMiMo/MiMo-V2.6-Pro-RL",[333,349,351],{"class":350},"sYu0t"," \\\n",[333,353,355,358,361],{"class":335,"line":354},2,[333,356,357],{"class":350},"  --tensor-parallel-size",[333,359,360],{"class":350}," 8",[333,362,351],{"class":350},[333,364,366,369],{"class":335,"line":365},3,[333,367,368],{"class":350},"  --trust-remote-code",[333,370,351],{"class":350},[333,372,374,377,380],{"class":335,"line":373},4,[333,375,376],{"class":350},"  --reasoning-parser",[333,378,379],{"class":343}," mimo",[333,381,351],{"class":350},[333,383,385,388,390],{"class":335,"line":384},5,[333,386,387],{"class":350},"  --tool-call-parser",[333,389,379],{"class":343},[333,391,351],{"class":350},[333,393,395],{"class":335,"line":394},6,[333,396,397],{"class":350},"  --enable-auto-tool-choice\n",[14,399,400,401,404],{},"Skip ",[291,402,403],{},"--tool-call-parser mimo"," and your agent's tool calls come back as plain text. The agent \"thinks\" it called the tool. Nothing happens. You spend an evening reading logs.",[316,406,408],{"id":407},"what-about-ollama","What about Ollama?",[14,410,411],{},"There's no official MiMo-V2.6-Pro model in the Ollama library, and at 573 GB Pro was never going to fit on a laptop. Community quantisations exist for Flash and the 9B distill, but they're unofficial. If you go that route, test tool calling with one simple tool before an agent depends on it.",[14,413,414],{},[46,415],{"alt":416,"src":417},"Three ways to run MiMo-V2.6-Pro: Xiaomi's API with model ID mimo-v2.6-pro (easiest), OpenRouter with xiaomi/mimo-v2.6-pro (one key), or self-hosting with --tool-call-parser mimo on 8x H200. Self-hosting Pro is datacentre-only.","/img/blog/mimo-v2-6-pro-agents-setup-cost-three-ways-to-run.jpg",[27,419,421],{"id":420},"running-mimo-v26-pro-in-betterclaw","Running MiMo-V2.6-Pro in BetterClaw",[14,423,424],{},"If you'd rather skip the server, BetterClaw lists Xiaomi as a model provider, alongside OpenRouter, Z.AI's GLM subscription and MiniMax. Pick the provider in the agent's Model tab, choose the model, paste your key, and save.",[14,426,427,428,430,431,435],{},"If a brand-new model hasn't reached the picker yet, the Custom model option takes any model ID the provider exposes, so ",[291,429,293],{}," works the day Xiaomi ships it. Our ",[266,432,434],{"href":433},"/docs","providers and models docs"," list everything supported.",[14,437,438],{},"The real win is per-agent models. Put MiMo-V2.6-Flash on your inbox triage agent, MiMo-V2.6-Pro on research, and keep Claude on the one agent that writes to customers. Switch any of them later without re-entering a key.",[14,440,441,442,448],{},"That's exactly why we built BetterClaw around bring-your-own-key. Plans start at $19 a month, you pay Xiaomi directly at their price with zero markup from us, and there's a 7-day money-back guarantee. ",[266,443,447],{"href":444,"rel":445},"https://app.betterclaw.io/sign-in",[446],"nofollow","Start with BetterClaw"," and test MiMo against your current model on real work.",[27,450,452],{"id":451},"the-footnote-anthropics-distillation-report","The footnote: Anthropic's distillation report",[14,454,455],{},"Here's what nobody tells you in the launch posts. Eleven days before MiMo-V2.6 shipped, Anthropic published a threat intelligence report naming seven China-based labs it says ran illicit distillation campaigns against Claude.",[14,457,458],{},"Xiaomi is one of them. Anthropic alleges that over about 20 days in March and April 2026, Xiaomi replayed user conversations and coding sessions from its own MiMo models to Claude, including through the OpenClaw and OpenCode coding tools, across more than 400,000 exchanges. Zhipu (the lab behind GLM) and MiniMax are named in the same report.",[14,460,461],{},"These are allegations from one company, not findings by a court or regulator, and press reports say China's Commerce Ministry has rejected them. The reported dates also come before the V2.6 release, and nothing public ties them to this specific model.",[14,463,464],{},"Our take: it doesn't change the MIT licence or the benchmarks. It may matter for your procurement rules. If your company has a policy about vendors named in allegations like these, it applies equally to GLM and MiniMax, so switching between the three doesn't change your answer.",[27,466,468],{"id":467},"should-you-use-mimo-v26-pro-for-agents","Should you use MiMo-V2.6-Pro for agents?",[14,470,471,474],{},[113,472,473],{},"Use it"," if your agents run in the background (research, reports, coding tasks, scheduled jobs) where cost per task matters more than speed. That's its sweet spot.",[14,476,477,480],{},[113,478,479],{},"Think twice"," if your agent chats with people in real time. At about 43 tokens per second plus long reasoning traces, users will notice the wait. DeepSeek V4.1 Flash measured about 214 tokens per second on the same index.",[14,482,483,486],{},[113,484,485],{},"Use Flash"," if volume matters more than the last few points of quality. It's within a few points of Pro on most of Xiaomi's agent benchmarks at about a third of the price.",[14,488,489,490,494],{},"And test before you switch. If you came over from self-hosting OpenClaw, our list of the ",[266,491,493],{"href":492},"/blog/cheapest-openclaw-ai-providers","cheapest AI providers for OpenClaw agents"," shows how much model choice moves the monthly bill.",[27,496,498],{"id":497},"the-bigger-shift","The bigger shift",[14,500,501],{},"A year ago, \"open-weight\" meant \"a solid step behind, but free.\" Now the best open model costs pennies per task and trails the closed frontier by about a dozen index points.",[14,503,504],{},"That gap is the only thing still worth paying a premium for. For everything below it, the question isn't which model is smartest. It's which one finishes your agent's job for the least money, at a speed your users will tolerate.",[14,506,507],{},"The best model for your agent is rarely the best model. It's the cheapest one that doesn't make mistakes you have to clean up.",[14,509,510,511,515],{},"If any of this resonated, give BetterClaw a try. Plans start at $19 a month for one agent, Pro is $49 for five, and every plan has a 7-day money-back guarantee. Bring MiMo, GLM, MiniMax, Claude or any of 28+ providers, and pay them directly with zero markup. ",[266,512,514],{"href":513},"/pricing","See full pricing",". We handle the infrastructure. You handle the interesting part.",[27,517,519],{"id":518},"frequently-asked-questions","Frequently Asked Questions",[316,521,30],{"id":522},"what-is-mimo-v26-pro-1",[14,524,525],{},"MiMo-V2.6-Pro is Xiaomi's open-weight flagship language model, released on 21 September 2026 under the MIT licence. It has 1.02 trillion total parameters with 42 billion active, a 1M-token context window, and accepts text, image, video and audio. Artificial Analysis scores it 46, the highest of any open-weight model it tracks.",[316,527,529],{"id":528},"how-does-mimo-v26-pro-compare-to-glm-53","How does MiMo-V2.6-Pro compare to GLM-5.3?",[14,531,532],{},"They're nearly tied on quality: 46 vs 45 on the Artificial Analysis Intelligence Index. MiMo-V2.6-Pro is far cheaper, at $0.435/$0.87 per million tokens against GLM-5.3's $1.40/$4.40, and about 15x cheaper per finished task on Artificial Analysis's measurements. GLM-5.3 is faster, at about 71 tokens per second against MiMo's 43.",[316,534,536],{"id":535},"how-do-i-use-mimo-v26-pro-with-tool-calling","How do I use MiMo-V2.6-Pro with tool calling?",[14,538,539,540,542,543,546,547,550],{},"Through Xiaomi's API or OpenRouter, tool calling works out of the box. If you self-host with vLLM, pass ",[291,541,403],{},", ",[291,544,545],{},"--reasoning-parser mimo"," and ",[291,548,549],{},"--enable-auto-tool-choice",", or tool calls come back as plain text. Test with a single simple tool before handing the agent real work.",[316,552,554],{"id":553},"how-much-does-mimo-v26-pro-cost","How much does MiMo-V2.6-Pro cost?",[14,556,557],{},"The API costs $0.435 per million input tokens, $0.87 per million output tokens and $0.0036 per million cached input tokens. For a typical agent doing about 300 tasks a month, that's roughly $3 in model usage before caching. MiMo-V2.6-Flash costs $0.14/$0.28.",[316,559,561],{"id":560},"is-mimo-v26-pro-safe-to-use-for-business-agents","Is MiMo-V2.6-Pro safe to use for business agents?",[14,563,564],{},"The weights are MIT-licensed and you can self-host them, which keeps data inside your own infrastructure. Through an API, your data goes to whichever provider serves it, so check their data policy. Anthropic has alleged Xiaomi distilled Claude, which China's Commerce Ministry rejects; if your company has vendor rules about that, apply them consistently, since GLM's and MiniMax's makers are named too.",[316,566,568],{"id":567},"can-i-run-mimo-v26-pro-with-ollama","Can I run MiMo-V2.6-Pro with Ollama?",[14,570,571],{},"Not practically. There's no official Ollama model for it, and the Pro checkpoint is about 573 GB. Community quantisations exist for MiMo-V2.6-Flash and the 9B distilled model, but they're unofficial, so test tool calling before relying on them.",[573,574,575],"style",{},"html pre.shiki code .s7eDp, html code.shiki .s7eDp{--shiki-default:#6F42C1}html pre.shiki code .sYBdl, html code.shiki .sYBdl{--shiki-default:#032F62}html pre.shiki code .sYu0t, html code.shiki .sYu0t{--shiki-default:#005CC5}html .default .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}",{"title":329,"searchDepth":354,"depth":354,"links":577},[578,579,580,581,585,586,587,588,589],{"id":29,"depth":354,"text":30},{"id":52,"depth":354,"text":53},{"id":74,"depth":354,"text":75},{"id":279,"depth":354,"text":280,"children":582},[583,584],{"id":318,"depth":365,"text":319},{"id":407,"depth":365,"text":408},{"id":420,"depth":354,"text":421},{"id":451,"depth":354,"text":452},{"id":467,"depth":354,"text":468},{"id":497,"depth":354,"text":498},{"id":518,"depth":354,"text":519,"children":590},[591,592,593,594,595,596],{"id":522,"depth":365,"text":30},{"id":528,"depth":365,"text":529},{"id":535,"depth":365,"text":536},{"id":553,"depth":365,"text":554},{"id":560,"depth":365,"text":561},{"id":567,"depth":365,"text":568},"Comparison","2026-10-05","MiMo V2.6 Pro tops open-weight models at $0.435/$0.87 per million tokens. Agent setup, tool calling, real costs vs GLM-5.3 and MiniMax M3.","md",false,"/img/blog/mimo-v2-6-pro-agents-setup-cost.jpg","MiMo-V2.6-Pro for AI Agents: setup and cost vs GLM and MiniMax, with a card showing an Intelligence Index of 46, $0.435 in and $0.87 out per million tokens, open weights under MIT.",512,1024,null,{},true,"/blog/mimo-v2-6-pro-agents-setup-cost","8 min read",{"title":5,"description":599},"blog/mimo-v2-6-pro-agents-setup-cost",[614,615,616,617,618,619,620,621],"mimo v2.6 pro","xiaomi mimo","mimo v2.6 pro benchmark","mimo v2.6 pro pricing","mimo vs glm","mimo v2.6 flash","mimo v2.6 pro tool calling","best open weights model 2026","amA4hJK1X2lUyPVrYDZ3I6vWZ0ECMefZQcL-2kkJbdo",[624,632,640],{"stem":625,"title":626,"description":627,"date":628,"image":629,"category":597,"redirected":601,"author":630,"readingTime":631},"blog/zanus-ai-vs-nvidia-dgx","Zanus AI Review: Private AI Servers vs NVIDIA DGX Spark and Station","Zanus AI review: what its private AI servers include, why pricing is quote-only, what the vendor won't publish, and when an NVIDIA DGX makes more sense.","2026-08-18","/img/blog/zanus-ai-vs-nvidia-dgx.jpg",{"name":7,"role":8,"avatar":9},"9 min read",{"stem":633,"title":634,"description":635,"date":636,"image":637,"category":597,"redirected":601,"author":638,"readingTime":639},"blog/dgx-spark-alternative","DGX Spark Alternatives 2026: 6 Cheaper Options From $0 to $3,099","DGX Spark now costs $6,950 and runs Linux only. Six cheaper alternatives ranked by price, from free Ollama and cloud APIs to Strix Halo PCs and Mac Studio M5.","2026-06-16","/img/blog/dgx-spark-alternative.jpg",{"name":7,"role":8,"avatar":9},"16 min read",{"stem":641,"title":642,"description":643,"date":644,"image":645,"category":597,"redirected":601,"author":646,"readingTime":647},"blog/openrouter-vs-direct-api-agents","OpenRouter vs Direct API: Is It Cheaper? Real Cost Math (2026)","OpenRouter pricing explained: the 5.5% fee, how BYOK removes it, free models, and the monthly spend where going direct to the API is cheaper.","2026-06-03","/img/blog/openrouter-vs-direct-api-agents.jpg",{"name":7,"role":8,"avatar":9},"14 min read",1791202883501]