[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"article-mimo-v25-pro-turns-agent-work-into-one-api-call-en":3,"article-related-mimo-v25-pro-turns-agent-work-into-one-api-call-en":30,"series-tools-072db467-c073-42e5-b723-5b1888304902":84},{"id":4,"slug":5,"title":6,"content":7,"summary":8,"source":9,"source_url":10,"author":11,"image_url":12,"cover_image":12,"category":13,"language":14,"translated_content":11,"related_article_id":15,"keywords":16,"key_takeaways":22,"views":26,"created_at":27,"published_at":28,"topic_cluster_id":29},"072db467-c073-42e5-b723-5b1888304902","mimo-v25-pro-turns-agent-work-into-one-api-call-en","MiMo-V2.5-Pro turns agent work into one API call","\u003Cp data-speakable=\"summary\">This shows how to wire MiMo-V2.5-Pro into \u003Ca href=\"\u002Ftag\u002Fagent\">agent\u003C\u002Fa> workflows and copy the setup.\u003C\u002Fp>\u003Cp>I've been messing with agent models long enough to know when a spec page is trying to sell me a feeling instead of a workflow. MiMo-V2.5-Pro on OpenRouter felt like that at first. Big context window, fancy \u003Ca href=\"\u002Ftag\u002Fbenchmark\">benchmark\u003C\u002Fa> names, a pile of promises about long-horizon tasks. Fine. I’ve seen plenty of models claim they can “do agentic work” and then fall apart the second you give them a real tool loop, a messy repo, or a request that needs restraint instead of enthusiasm.\u003C\u002Fp>\u003Cp>What annoyed me here was simpler: I didn’t want another model that sounds smart in a demo. I wanted to know whether this thing is actually worth wiring into a production-ish agent stack, what the pricing really means, and whether OpenRouter’s routing and provider setup makes the whole thing easier or just adds another layer to debug. The page gives enough hints to make that judgment, but only if you strip away the marketing and read it like an operator.\u003C\u002Fp>\u003Cp>The source that triggered this breakdown is the OpenRouter model page for \u003Ca href=\"https:\u002F\u002Fopenrouter.ai\u002Fxiaomi\u002Fmimo-v2.5-pro\">MiMo-V2.5-Pro\u003C\u002Fa>. OpenRouter also links the surrounding docs for \u003Ca href=\"https:\u002F\u002Fopenrouter.ai\u002Fdocs\">API usage\u003C\u002Fa>, \u003Ca href=\"https:\u002F\u002Fopenrouter.ai\u002Fdocs\u002Fquick-start\">quick start\u003C\u002Fa>, and \u003Ca href=\"https:\u002F\u002Fopenrouter.ai\u002Fdocs\u002Frouting\">routing modes\u003C\u002Fa>, which matter more here than the model name itself. I’m not pulling in outside benchmark claims beyond what the page states: ClawEval, GDPVal, and \u003Ca href=\"\u002Ftag\u002Fswe-bench\">SWE-bench\u003C\u002Fa> Pro, plus the listed pricing of $0.435 per million input tokens and $0.87 per million output tokens.\u003C\u002Fp>\u003Ch2>It’s not a chat toy, it’s an agent budget line\u003C\u002Fh2>\u003Cblockquote>MiMo-V2.5-Pro is Xiaomi’s flagship model, delivering strong performance in general agentic capabilities, complex software engineering, and long-horizon tasks, with top rankings on benchmarks such as ClawEval, GDPVal, and SWE-bench Pro. It can independently and autonomously complete professional tasks that would take human experts days or weeks, involving more than a thousand tool calls.\u003C\u002Fblockquote>\u003Cp>What this actually means is: Xiaomi is not positioning this as a “better chatbot.” They’re saying, “Use me when the job is a sequence of decisions, tool calls, retries, and state tracking.” That’s a very different promise. If a model is good at one-shot answers, great. If it can survive a thousand tool calls without wandering off, that’s the kind of thing you use for multi-step coding agents, research pipelines, or ops workflows that need persistence.\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781320696191-vzza.png\" alt=\"MiMo-V2.5-Pro turns agent work into one API call\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>I’ve run into this exact split before. A model can ace a short benchmark and still be a pain in a repo because it overcommits too early, forgets constraints, or keeps rewriting the plan every turn. The phrase “long-horizon tasks” is the part I care about. That’s where most agents die. They don’t fail loudly. They slowly drift until the output is technically responsive and practically useless.\u003C\u002Fp>\u003Cp>How to apply it: treat MiMo-V2.5-Pro like a worker model, not a brainstorm partner. Give it a task boundary, a tool policy, and a stop condition. If your stack has planner\u002Fexecutor separation, this is the executor side. If you only have one model in the loop, you’ll want a very strict system prompt and a hard cap on tool retries.\u003C\u002Fp>\u003Cul>\u003Cli>Use it for tasks with explicit state: code changes, file edits, ticket triage, structured research.\u003C\u002Fli>\u003Cli>Don’t use it for open-ended ideation unless you want a confident mess.\u003C\u002Fli>\u003Cli>Make the success criteria machine-checkable whenever possible.\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>The 1M context window is the real flex\u003C\u002Fh2>\u003Cp>The page says MiMo-V2.5-Pro has a context length of up to 1M. That is not a vanity number. That changes how I think about retrieval, memory, and chunking. With a million tokens, the model can hold a ridiculous amount of source material in one pass, which means fewer retrieval hops and fewer chances to lose context between calls.\u003C\u002Fp>\u003Cp>But I’ve also learned not to worship giant context windows. A huge window does not magically make the model attentive to everything you stuffed into it. It just means you have more room to be sloppy. If you dump a whole codebase into the prompt and expect the model to behave like a senior engineer with perfect recall, you’re going to get burned.\u003C\u002Fp>\u003Cp>What this actually means is: the 1M context is best used to reduce fragmentation, not to avoid design. I’d use it for large spec bundles, long issue threads, multi-file refactors, or agent runs where prior decisions need to stay visible. I would not use it as permission to stop curating input.\u003C\u002Fp>\u003Cp>I ran into this when I tested long task chains on models with smaller windows. The constant summarization step became the failure point. Every time the agent summarized, nuance leaked out. A larger context window cuts down that leak. That alone can be worth a lot if your workflow depends on keeping constraints intact over many turns.\u003C\u002Fp>\u003Cp>How to apply it: keep a compact “task ledger” inside the prompt. Don’t rely on the model to infer the current state from a wall of text. Put the goal, constraints, completed steps, and remaining steps in a dedicated block that gets refreshed each turn.\u003C\u002Fp>\u003Cul>\u003Cli>Use a running state section: goal, constraints, done, next, blocked.\u003C\u002Fli>\u003Cli>Keep raw artifacts in context only when they are directly relevant.\u003C\u002Fli>\u003Cli>For code tasks, include filenames and diffs, not the whole repo unless needed.\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>OpenRouter matters because provider chaos is real\u003C\u002Fh2>\u003Cp>OpenRouter says different companies host the same model, and it routes requests based on the mode you pick: Balanced, Nitro, or Exacto. That sounds like plumbing, but it’s the part that decides whether your agent feels stable or flaky. I care less about the model page and more about whether the routing layer gives me predictable behavior when one provider coughs.\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781320692543-gp74.png\" alt=\"MiMo-V2.5-Pro turns agent work into one API call\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>What this actually means is: OpenRouter is acting like the control plane. The model is the payload. If you’ve ever had one provider go weird at 2 a.m. and your “production” agent suddenly start timing out, you already know why this matters. OpenRouter also says it monitors providers continuously and retries on the next-best provider when one errors. That’s not glamorous, but it’s the kind of thing that keeps a workflow from face-planting on a bad upstream hour.\u003C\u002Fp>\u003Cp>I’ve had enough model integrations where the API was fine in testing and then became a support ticket factory in production because the hosting layer changed behavior. Routing modes are the difference between “I want the cheapest workable path” and “I need the same provider every time because I’m debugging.”\u003C\u002Fp>\u003Cp>How to apply it: pick a routing mode based on the job, not the mood.\u003C\u002Fp>\u003Cul>\u003Cli>\u003Cstrong>Balanced\u003C\u002Fstrong> when you want price and speed to trade off automatically.\u003C\u002Fli>\u003Cli>\u003Cstrong>Nitro\u003C\u002Fstrong> when latency matters more than cost.\u003C\u002Fli>\u003Cli>\u003Cstrong>Exacto\u003C\u002Fstrong> when you need one fixed provider for repeatability.\u003C\u002Fli>\u003C\u002Ful>\u003Cp>If you’re evaluating the model, start with Exacto so you can isolate model behavior from provider noise. Once it’s stable, move to Balanced if cost matters. That sequence saves time and makes your debugging sane.\u003C\u002Fp>\u003Ch2>The pricing is low enough to invite abuse, so put guardrails on it\u003C\u002Fh2>\u003Cp>The listed price is $0.435 per million input tokens and $0.87 per million output tokens. That is cheap enough that people will immediately start imagining giant agent loops, long prompts, and “let it think forever” workflows. I get the temptation. I also know how quickly cheap output becomes expensive when your agent starts rambling, retrying, or generating unnecessary intermediate text.\u003C\u002Fp>\u003Cp>What this actually means is: pricing only helps if your control loop is disciplined. A model like this becomes attractive when you can amortize the context cost across a long task. If your app burns tokens on repetitive prompt scaffolding or verbose tool chatter, you’re throwing away the advantage.\u003C\u002Fp>\u003Cp>I’ve seen teams lower model cost and then accidentally raise total spend because they removed one bottleneck but kept the same loose orchestration. The model was cheaper, sure. The workflow was not. The right move is to reduce useless tokens first, then pick the model.\u003C\u002Fp>\u003Cp>How to apply it: design for \u003Ca href=\"\u002Ftag\u002Ftoken\">token\u003C\u002Fa> economy before you scale up usage. Keep tool outputs structured. Trim repeated instructions. Cache stable context where possible. And if you’re using prompt caching, remember OpenRouter says effective price can be 60–80% cheaper for repeated context. That’s the kind of detail that can make a real difference in a loop that reuses the same task frame.\u003C\u002Fp>\u003Cp>My rule of thumb: if a step doesn’t change the decision, don’t keep paying for it.\u003C\u002Fp>\u003Ch2>The benchmarks tell you where to trust it, not where to worship it\u003C\u002Fh2>\u003Cp>OpenRouter lists ClawEval, GDPVal, and SWE-bench Pro as the headline benchmarks. I’m happy they named them, because benchmark names are usually where model pages get vague and hand-wavy. Here, the signal is pretty clear: they want you to think about agent execution, software engineering, and longer work streams.\u003C\u002Fp>\u003Cp>What this actually means is: I would read those benchmarks as a workload hint. Not as a blanket claim that the model is best at everything. If your use case is coding assistance, repo navigation, or multi-step task completion, these scores are relevant. If your use case is creative writing or customer support tone control, they’re much less informative.\u003C\u002Fp>\u003Cp>I’ve learned to ask one blunt question whenever a model page throws benchmark names at me: does this benchmark resemble my actual failure mode? If the answer is no, I stop caring. A model can look great on a leaderboard and still make a mess of your particular task shape.\u003C\u002Fp>\u003Cp>How to apply it: map each benchmark to a real internal task. For example:\u003C\u002Fp>\u003Cul>\u003Cli>Software engineering benchmark → code review, refactor, test generation.\u003C\u002Fli>\u003Cli>Agentic benchmark → ticket resolution, tool chaining, stateful workflows.\u003C\u002Fli>\u003Cli>Long-horizon benchmark → research synthesis, multi-step ops tasks.\u003C\u002Fli>\u003C\u002Ful>\u003Cp>If the benchmark doesn’t resemble your workload, don’t let it drive the decision. Use it as a hint, then test on your \u003Ca href=\"\u002Fnews\u002Fanthropic-ai-building-ai-recursive-self-improvement-en\">own data\u003C\u002Fa>.\u003C\u002Fp>\u003Ch2>OpenAI-compatible API means less glue code, which I appreciate\u003C\u002Fh2>\u003Cp>OpenRouter says its API is \u003Ca href=\"\u002Ftag\u002Fopenai\">OpenAI\u003C\u002Fa>-compatible and that most SDKs work by just swapping the base URL. Good. That’s the kind of boring sentence I like. It means I can point existing clients at the new endpoint and keep moving instead of rewriting half my integration layer for a branding exercise.\u003C\u002Fp>\u003Cp>What this actually means is: the model choice changes, the client shape mostly doesn’t. That lowers the cost of experimentation. You can swap MiMo-V2.5-Pro into an existing agent stack, compare behavior, and keep the rest of the plumbing stable. That matters when you’re evaluating models across multiple providers and don’t want your benchmark harness to become the thing you’re testing.\u003C\u002Fp>\u003Cp>I ran into this when I was comparing models in a tool-using workflow. The moment the API shape diverged, I stopped learning about the model and started debugging adapters. OpenAI-compatible endpoints are not sexy, but they keep the evaluation honest.\u003C\u002Fp>\u003Cp>How to apply it: keep one thin provider adapter in your codebase and point it at OpenRouter’s base URL. Then keep model selection as a config value, not a code fork. That makes it easy to swap between MiMo-V2.5-Pro and something else without touching the rest of the agent.\u003C\u002Fp>\u003Cp>Here’s the part people skip: compatibility is only useful if your prompts, tools, and output parsing are already clean. If your integration is sloppy, a compatible API just helps you move the mess faster.\u003C\u002Fp>\u003Ch2>The template you can copy\u003C\u002Fh2>\u003Cpre>\u003Ccode>## MiMo-V2.5-Pro agent setup via OpenRouter\n\n### When I use this\nUse MiMo-V2.5-Pro for long-horizon, tool-heavy tasks:\n- codebase refactors\n- issue triage\n- research synthesis\n- multi-step agent workflows\n- tasks that need a large shared context\n\n### Provider strategy\n- Start with `Exacto` while testing so provider behavior stays fixed.\n- Move to `Balanced` for production if cost and speed both matter.\n- Use `Nitro` only when latency matters more than repeatability.\n\n### Prompt structure\ntext\nYou are a task executor.\n\nGoal:\n{one-sentence goal}\n\nConstraints:\n- Do not change unrelated files\n- Prefer minimal diffs\n- Ask before destructive actions\n- Keep outputs structured\n\nState:\n- Done: {completed steps}\n- Next: {next step}\n- Blocked: {open issues}\n\nTools:\n- Use tools only when they reduce uncertainty\n- Stop when the task is complete\n- Report failures with exact error text\n\nOutput format:\n1. Summary\n2. Actions taken\n3. Files changed\n4. Remaining risks\n\n\n### OpenRouter request example\n\n{\n  \"model\": \"xiaomi\u002Fmimo-v2.5-pro\",\n  \"messages\": [\n    {\"role\": \"system\", \"content\": \"You are a task executor...\"},\n    {\"role\": \"user\", \"content\": \"{task}\"}\n  ],\n  \"temperature\": 0.2\n}\n\n\n### Practical guardrails\n- Keep a running task ledger in the prompt.\n- Use structured tool outputs.\n- Refresh only the state that changed.\n- Cap retries for tool calls.\n- Trim repeated instructions so you do not waste context.\n\n### Evaluation checklist\n- Did the model stay on task after multiple tool calls?\n- Did it preserve constraints across turns?\n- Did it complete the job without unnecessary replanning?\n- Did provider switching change behavior?\n- Did the output stay useful after long context usage?\n\n### Copy-paste starter config\nyaml\nprovider: openrouter\nmodel: xiaomi\u002Fmimo-v2.5-pro\nrouting_mode: exacto\ntemperature: 0.2\nmax_retries: 2\ncontext_policy: keep_state_compact\noutput_style: structured\n\n\u003C\u002Fcode>\u003C\u002Fpre>\u003Cp>That template is mine, not OpenRouter’s. I built it from the model page, the routing docs, and the usual pain of trying to keep agent workflows from turning into token bonfires. The model slug is from \u003Ca href=\"https:\u002F\u002Fopenrouter.ai\u002Fxiaomi\u002Fmimo-v2.5-pro\">OpenRouter’s MiMo-V2.5-Pro page\u003C\u002Fa>; the rest is the copy-ready part you can adapt to your own stack.\u003C\u002Fp>\u003Cp>If you want to compare against the source directly, start with the model page and then check \u003Ca href=\"https:\u002F\u002Fopenrouter.ai\u002Fdocs\">OpenRouter docs\u003C\u002Fa>, \u003Ca href=\"https:\u002F\u002Fopenrouter.ai\u002Fdocs\u002Fquick-start\">quick start\u003C\u002Fa>, and \u003Ca href=\"https:\u002F\u002Fopenrouter.ai\u002Fdocs\u002Frouting\">routing\u003C\u002Fa>. I’m not pretending this article is the source of truth. It’s my read on how to actually use the thing without getting hypnotized by benchmark talk.\u003C\u002Fp>","A practical breakdown of Xiaomi’s MiMo-V2.5-Pro on OpenRouter, plus a copy-ready setup for agentic tool workflows.","openrouter.ai","https:\u002F\u002Fopenrouter.ai\u002Fxiaomi\u002Fmimo-v2.5-pro",null,"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781320696191-vzza.png","tools","en","1f8c57f4-698a-42a5-80e8-85cf7cb915d6",[17,18,19,20,21],"MiMo-V2.5-Pro","OpenRouter","agentic workflows","SWE-bench Pro","API pricing",[23,24,25],"Treat MiMo-V2.5-Pro as a worker model for long, tool-heavy tasks.","Use OpenRouter routing modes to separate testing from production behavior.","Keep prompts and state compact so the low token price doesn’t get wasted.",0,"2026-06-13T03:17:48.852122+00:00","2026-06-13T03:17:48.843+00:00","7d5bcbd3-cee8-4d2c-9da7-160cf0cf7a46",{"tags":31,"relatedLang":43,"relatedPosts":47},[32,34,36,38,41],{"name":18,"slug":33},"openrouter",{"name":17,"slug":35},"mimo-v25-pro",{"name":21,"slug":37},"api-pricing",{"name":39,"slug":40},"SWE-Bench Pro","swe-bench-pro",{"name":19,"slug":42},"agentic-workflows",{"id":15,"slug":44,"title":45,"language":46},"mimo-v25-pro-turns-agent-work-into-one-api-call-zh","MiMo-V2.5-Pro 把 agent 工作變成一個 API","zh",[48,54,60,66,72,78],{"id":49,"slug":50,"title":51,"cover_image":52,"image_url":52,"created_at":53,"category":13},"4543733b-22c2-4f7d-a115-038d4c542cd2","microsoft-build-2026-agents-into-systems-en","Microsoft Build 2026 turns agents into systems","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781329728835-fcsi.png","2026-06-13T05:48:22.073787+00:00",{"id":55,"slug":56,"title":57,"cover_image":58,"image_url":58,"created_at":59,"category":13},"1e01a6f4-3f54-407d-a775-f9147de8b911","8-cursor-alternatives-that-fit-how-you-work-en","8 Cursor alternatives that fit how you work","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781312598893-8ox5.png","2026-06-13T01:02:56.581668+00:00",{"id":61,"slug":62,"title":63,"cover_image":64,"image_url":64,"created_at":65,"category":13},"1a47f4ce-884c-4c5f-b5e1-2b117799fcce","ollama-default-local-ai-layer-en","Ollama is becoming the default local AI layer","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781310763401-cv9v.png","2026-06-13T00:32:16.812947+00:00",{"id":67,"slug":68,"title":69,"cover_image":70,"image_url":70,"created_at":71,"category":13},"105a8ad6-b183-47dd-9a1d-536c6ba89371","songscription-turns-piano-audio-into-editable-sheet-music-en","Songscription turns piano audio into editable sheet music","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781288304586-nos2.png","2026-06-12T18:17:56.189061+00:00",{"id":73,"slug":74,"title":75,"cover_image":76,"image_url":76,"created_at":77,"category":13},"9732ff57-20bf-4d96-aab9-6433cf93a6fc","ios-27-turns-apple-music-into-smarter-player-en","iOS 27 turns Apple Music into a smarter player","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781287409882-9rve.png","2026-06-12T18:02:58.670536+00:00",{"id":79,"slug":80,"title":81,"cover_image":82,"image_url":82,"created_at":83,"category":13},"0abd6ddf-cb83-497b-a70b-9b559356970f","kubernetes-turns-container-chaos-into-operations-en","Kubernetes turns container chaos into operations","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781285606704-qzol.png","2026-06-12T17:32:53.646476+00:00",[85,90,95,100,105,110,115,120,125,130],{"id":86,"slug":87,"title":88,"created_at":89},"8008f1a9-7a00-4bad-88c9-3eedc9c6b4b1","surepath-ai-mcp-policy-controls-en","SurePath AI's New MCP Policy Controls Enhance AI Security","2026-03-26T01:26:52.222015+00:00",{"id":91,"slug":92,"title":93,"created_at":94},"27e39a8f-b65d-4f7b-a875-859e2b210156","mcp-standard-ai-tools-2026-en","MCP Standard in 2026: Integrating AI Tools","2026-03-26T01:27:43.127519+00:00",{"id":96,"slug":97,"title":98,"created_at":99},"165f9a19-c92d-46ba-b3f0-7125f662921d","rag-2026-transforming-enterprise-ai-en","How RAG in 2026 is Transforming Enterprise AI","2026-03-26T01:28:11.485236+00:00",{"id":101,"slug":102,"title":103,"created_at":104},"6a2a8e6e-b956-49d8-be12-cc47bdc132b2","mastering-ai-prompts-2026-guide-en","Mastering AI Prompts: A 2026 Guide for Developers","2026-03-26T01:29:07.835148+00:00",{"id":106,"slug":107,"title":108,"created_at":109},"3ab2c67e-4664-4c67-a013-687a2f605814","garry-tan-open-sources-claude-code-toolkit-en","Garry Tan Open-Sources a Claude Code Toolkit","2026-03-26T08:26:20.245934+00:00",{"id":111,"slug":112,"title":113,"created_at":114},"66a7cbf8-7e76-41d4-9bbf-eaca9761bf69","github-ai-projects-to-watch-in-2026-en","20 GitHub AI Projects to Watch in 2026","2026-03-26T08:28:09.752027+00:00",{"id":116,"slug":117,"title":118,"created_at":119},"9f332fda-eace-448a-a292-2283951eee71","practical-github-guide-learning-ml-2026-en","A Practical GitHub Guide to Learning ML in 2026","2026-03-27T01:16:50.125678+00:00",{"id":121,"slug":122,"title":123,"created_at":124},"1b1f637d-0f4d-42bd-974b-07b53829144d","aiml-2026-student-ai-ml-lab-repo-review-en","AIML-2026 Is a Bare-Bones Student Lab Repo","2026-03-27T01:21:51.661231+00:00",{"id":126,"slug":127,"title":128,"created_at":129},"6d1bf3f6-e191-4d30-b55b-8a0722fa6afe","ai-trending-github-repos-and-research-feeds-en","AI Trending Tracks Repos and Research Feeds","2026-03-27T01:31:35.709532+00:00",{"id":131,"slug":132,"title":133,"created_at":134},"010539a1-4c3a-4bd3-937a-26616422ee0d","awesome-ai-for-science-research-tools-map-en","Awesome AI for Science Is Becoming a Real Research Map","2026-03-27T01:46:50.89513+00:00"]