[{"data":1,"prerenderedAt":220},["ShallowReactive",2],{"blog-local-ai":3,"surround-local-ai":186,"i-tabler:chevron-down":191,"i-tabler:menu-2":196,"i-tabler:rocket":198,"i-tabler:arrow-right":200,"i-tabler:mail":202,"i-tabler:brand-discord":204,"i-tabler:brand-x":206,"i-tabler:brand-youtube":208,"i-tabler:brand-linkedin":210,"i-tabler:brand-facebook":212,"i-tabler:brand-instagram":214,"i-tabler:brand-tiktok":216,"i-tabler:arrow-up-right":218},{"id":4,"title":5,"author":6,"body":7,"canonical":168,"date":169,"description":170,"extension":171,"image":14,"meta":172,"navigation":173,"ogImage":168,"path":174,"seo":175,"stem":176,"tags":177,"twitterCard":184,"__hash__":185},"blog/blog/local-ai.md","Local AI's Open-Weight Resurgence","Nancy Parajuli",{"type":8,"value":9,"toc":156},"minimark",[10,15,19,22,25,30,33,36,39,43,46,49,52,55,58,62,65,68,72,75,78,81,84,88,91,94,98,103,106,109,112,123,126,131],[11,12],"blog-image",{"alt":5,"caption":13,"src":14},"Open-weight models are making local and self-hosted AI practical for real workflows.","/blog/local-ai/local-ai-preview.png",[16,17,18],"p",{},"Local AI really hit its stride in early 2025 when DeepSeek-R1 proved that open models could hold their own on serious reasoning tasks.",[16,20,21],{},"Afterward, the progress felt uneven. Cloud models took off, and local options often seemed about six months behind whenever the work got difficult. Now, that gap is closing again.",[16,23,24],{},"A new wave of open-weight models, led by Z.ai's GLM-5.2, is making local and self-hosted AI feel usable for coding, long-document work, and agent workflows.",[26,27,29],"h2",{"id":28},"the-gap-did-not-close-overnight","The Gap Did Not Close Overnight",[16,31,32],{},"DeepSeek-R1 made open reasoning feel real. For the first time, you could see the thoughts behind an AI model's response. But once that hype settled, progress in the West for open-weight models felt slow, especially compared to the pace of frontier cloud models that were being developed and that once again created all of the buzz.",[16,34,35],{},"Meanwhile, Chinese AI labs kept pushing open-weight development hard. DeepSeek, Qwen, Kimi, MiniMax, and now Z.ai's GLM-5.2 have all kept momentum going for open models.",[16,37,38],{},"For a while, local models were becoming more useful for basic tasks, but still playing catch-up, especially in terms of more complex workflows. GLM-5.2 is one of the clearest signs yet that the gap is closing between state-of-the-art proprietary models and open weights.",[26,40,42],{"id":41},"glm-52-raises-the-bar","GLM-5.2 Raises The Bar",[16,44,45],{},"GLM-5.2 stands out because it combines scale, openness, and strong performance.",[16,47,48],{},"Z.ai says the model has a 1M-token context window, 744B total parameters, 40B active parameters, and flexible thinking, all under an MIT license.",[16,50,51],{},"Artificial Analysis ranked GLM-5.2 as the top open-weight model on its Intelligence Index v4.1, showing particularly strong results on agentic knowledge work.",[16,53,54],{},"Most people will not be able to run the full model on a basic laptop because of how large it is, but having open weights is a huge deal because teams can host it, test it, and build around it without being locked into an expensive online provider.",[16,56,57],{},"We have officially entered the era of high stress and focus around token usage. Tokens are becoming a scary word, and with open-weight models, that stress has some relief.",[26,59,61],{"id":60},"real-world-reactions-are-driving-the-shift","Real-World Reactions Are Driving The Shift",[16,63,64],{},"GLM-5.2 is getting real attention because people building with AI are comparing it directly against top closed models.",[16,66,67],{},"Alex Finn said GLM-5.2 is comparable to Opus 4.8, framing local models as a real step toward independence. That's huge!",[69,70],"x-post",{"url":71},"https://x.com/AlexFinn/status/2067281289926103067",[16,73,74],{},"Nutlope put GLM-5.2 up against Opus 4.8 on a design task and pointed out the cost difference, noting that GLM gave them a strong result for a fraction of the price.",[69,76],{"url":77},"https://x.com/nutlope/status/2067313679951941686",[16,79,80],{},"Mitchell Hashimoto summed up the mood well. Local models have gone from feeling weak to being genuinely useful, even if they still have a way to go before they hit the highest bar. Open models are getting close!",[69,82],{"url":83},"https://x.com/mitchellh/status/2066960258304782598",[26,85,87],{"id":86},"open-models-are-where-privacy-lies","Open Models Are Where Privacy Lies",[16,89,90],{},"If privacy and security are top of mind, then you've been waiting for open models to catch up.",[16,92,93],{},"Open-weight models let teams keep control of their data, manage spend, and pick the model that fits the job. For organizations and people handling sensitive work, cloud models may never have been a viable option. But now, they are looking at privacy-first solutions that deliver results.",[26,95,97],{"id":96},"a-practical-local-workflow-in-msty","A Practical Local Workflow In Msty",[11,99],{"alt":100,"caption":101,"src":102},"Split Chat in Msty Studio - comparing multiple models","Split Chat in Msty Studio makes it easy to test the same prompt across models, from GLM-5.2 through an API to smaller local models like Qwen.","/blog/local-ai/ss-1.webp",[16,104,105],{},"Msty AI has been creating privacy-first and local-first products since day one.",[16,107,108],{},"In Msty Studio and Msty Claw, you can discover, install, and manage local models from within the apps, with support for runtimes like Local AI, LLaMA.cpp, and MLX.",[16,110,111],{},"You can also use open-weight models through their APIs. For instance, GLM-5.2 can be used in Msty Studio and Msty Claw by adding an API key, while local models remain available for private or cost-sensitive tasks.",[16,113,114,115,122],{},"To learn more about installing, importing, and running local models in Msty Studio, see the ",[116,117,121],"a",{"href":118,"rel":119},"https://docs.msty.ai/studio/managing-models/local-models",[120],"nofollow","Local Models guide",".",[16,124,125],{},"Simple. Private. Powerful.",[127,128,130],"h3",{"id":129},"sources","Sources",[132,133,134,142,149],"ul",{},[135,136,137],"li",{},[116,138,141],{"href":139,"rel":140},"https://z.ai/blog/glm-5.2",[120],"Source: Z.ai GLM-5.2 announcement",[135,143,144],{},[116,145,148],{"href":146,"rel":147},"https://huggingface.co/zai-org/GLM-5.2",[120],"Source: Z.ai GLM-5.2 model card",[135,150,151],{},[116,152,155],{"href":153,"rel":154},"https://artificialanalysis.ai/articles/glm-5-2-is-the-new-leading-open-weights-model-on-the-artificial-analysis-intelligence-index",[120],"Source: Artificial Analysis on GLM-5.2",{"title":157,"searchDepth":158,"depth":158,"links":159},"",2,[160,161,162,163,164],{"id":28,"depth":158,"text":29},{"id":41,"depth":158,"text":42},{"id":60,"depth":158,"text":61},{"id":86,"depth":158,"text":87},{"id":96,"depth":158,"text":97,"children":165},[166],{"id":129,"depth":167,"text":130},3,null,"20260623","Open-weight models are becoming practical for coding, long-document work, and agent workflows. GLM-5.2 shows that local and self-hosted AI is closing the gap with frontier cloud models.","md",{},true,"/blog/local-ai",{"title":5,"description":170},"blog/local-ai",[178,179,180,181,182,183],"local-ai","open-source-models","glm-5.2","msty","ai-models","self-hosted","summary_large_image","yfdZg27zKNUME-H-fuLoIt6VHvr41hTHnuM7NHptsHg",[168,187],{"title":188,"path":189,"stem":190,"children":-1},"Make Msty Claw Work Your Way with Extensions","/blog/extensions","blog/extensions",{"left":192,"top":192,"width":193,"height":193,"rotate":192,"vFlip":194,"hFlip":194,"body":195},0,24,false,"\u003Cpath fill=\"none\" stroke=\"currentColor\" stroke-linecap=\"round\" stroke-linejoin=\"round\" stroke-width=\"2\" d=\"m6 9l6 6l6-6\"/>",{"left":192,"top":192,"width":193,"height":193,"rotate":192,"vFlip":194,"hFlip":194,"body":197},"\u003Cpath fill=\"none\" stroke=\"currentColor\" stroke-linecap=\"round\" stroke-linejoin=\"round\" stroke-width=\"2\" d=\"M4 6h16M4 12h16M4 18h16\"/>",{"left":192,"top":192,"width":193,"height":193,"rotate":192,"vFlip":194,"hFlip":194,"body":199},"\u003Cg fill=\"none\" stroke=\"currentColor\" stroke-linecap=\"round\" stroke-linejoin=\"round\" stroke-width=\"2\">\u003Cpath d=\"M4 13a8 8 0 0 1 7 7a6 6 0 0 0 3-5a9 9 0 0 0 6-8a3 3 0 0 0-3-3a9 9 0 0 0-8 6a6 6 0 0 0-5 3\"/>\u003Cpath d=\"M7 14a6 6 0 0 0-3 6a6 6 0 0 0 6-3m4-8a1 1 0 1 0 2 0a1 1 0 1 0-2 0\"/>\u003C/g>",{"left":192,"top":192,"width":193,"height":193,"rotate":192,"vFlip":194,"hFlip":194,"body":201},"\u003Cpath fill=\"none\" stroke=\"currentColor\" stroke-linecap=\"round\" stroke-linejoin=\"round\" stroke-width=\"2\" d=\"M5 12h14m-6 6l6-6m-6-6l6 6\"/>",{"left":192,"top":192,"width":193,"height":193,"rotate":192,"vFlip":194,"hFlip":194,"body":203},"\u003Cg fill=\"none\" stroke=\"currentColor\" stroke-linecap=\"round\" stroke-linejoin=\"round\" stroke-width=\"2\">\u003Cpath d=\"M3 7a2 2 0 0 1 2-2h14a2 2 0 0 1 2 2v10a2 2 0 0 1-2 2H5a2 2 0 0 1-2-2z\"/>\u003Cpath d=\"m3 7l9 6l9-6\"/>\u003C/g>",{"left":192,"top":192,"width":193,"height":193,"rotate":192,"vFlip":194,"hFlip":194,"body":205},"\u003Cg fill=\"none\" stroke=\"currentColor\" stroke-linecap=\"round\" stroke-linejoin=\"round\" stroke-width=\"2\">\u003Cpath d=\"M8 12a1 1 0 1 0 2 0a1 1 0 0 0-2 0m6 0a1 1 0 1 0 2 0a1 1 0 0 0-2 0\"/>\u003Cpath d=\"M15.5 17c0 1 1.5 3 2 3c1.5 0 2.833-1.667 3.5-3c.667-1.667.5-5.833-1.5-11.5c-1.457-1.015-3-1.34-4.5-1.5l-.972 1.923a11.9 11.9 0 0 0-4.053 0L9 4c-1.5.16-3.043.485-4.5 1.5c-2 5.667-2.167 9.833-1.5 11.5c.667 1.333 2 3 3.5 3c.5 0 2-2 2-3\"/>\u003Cpath d=\"M7 16.5c3.5 1 6.5 1 10 0\"/>\u003C/g>",{"left":192,"top":192,"width":193,"height":193,"rotate":192,"vFlip":194,"hFlip":194,"body":207},"\u003Cpath fill=\"none\" stroke=\"currentColor\" stroke-linecap=\"round\" stroke-linejoin=\"round\" stroke-width=\"2\" d=\"m4 4l11.733 16H20L8.267 4zm0 16l6.768-6.768m2.46-2.46L20 4\"/>",{"left":192,"top":192,"width":193,"height":193,"rotate":192,"vFlip":194,"hFlip":194,"body":209},"\u003Cg fill=\"none\" stroke=\"currentColor\" stroke-linecap=\"round\" stroke-linejoin=\"round\" stroke-width=\"2\">\u003Cpath d=\"M2 8a4 4 0 0 1 4-4h12a4 4 0 0 1 4 4v8a4 4 0 0 1-4 4H6a4 4 0 0 1-4-4z\"/>\u003Cpath d=\"m10 9l5 3l-5 3z\"/>\u003C/g>",{"left":192,"top":192,"width":193,"height":193,"rotate":192,"vFlip":194,"hFlip":194,"body":211},"\u003Cg fill=\"none\" stroke=\"currentColor\" stroke-linecap=\"round\" stroke-linejoin=\"round\" stroke-width=\"2\">\u003Cpath d=\"M8 11v5m0-8v.01M12 16v-5m4 5v-3a2 2 0 1 0-4 0\"/>\u003Cpath d=\"M3 7a4 4 0 0 1 4-4h10a4 4 0 0 1 4 4v10a4 4 0 0 1-4 4H7a4 4 0 0 1-4-4z\"/>\u003C/g>",{"left":192,"top":192,"width":193,"height":193,"rotate":192,"vFlip":194,"hFlip":194,"body":213},"\u003Cpath fill=\"none\" stroke=\"currentColor\" stroke-linecap=\"round\" stroke-linejoin=\"round\" stroke-width=\"2\" d=\"M7 10v4h3v7h4v-7h3l1-4h-4V8a1 1 0 0 1 1-1h3V3h-3a5 5 0 0 0-5 5v2z\"/>",{"left":192,"top":192,"width":193,"height":193,"rotate":192,"vFlip":194,"hFlip":194,"body":215},"\u003Cg fill=\"none\" stroke=\"currentColor\" stroke-linecap=\"round\" stroke-linejoin=\"round\" stroke-width=\"2\">\u003Cpath d=\"M4 8a4 4 0 0 1 4-4h8a4 4 0 0 1 4 4v8a4 4 0 0 1-4 4H8a4 4 0 0 1-4-4z\"/>\u003Cpath d=\"M9 12a3 3 0 1 0 6 0a3 3 0 0 0-6 0m7.5-4.5v.01\"/>\u003C/g>",{"left":192,"top":192,"width":193,"height":193,"rotate":192,"vFlip":194,"hFlip":194,"body":217},"\u003Cpath fill=\"none\" stroke=\"currentColor\" stroke-linecap=\"round\" stroke-linejoin=\"round\" stroke-width=\"2\" d=\"M21 7.917v4.034A9.95 9.95 0 0 1 16 10v4.5a6.5 6.5 0 1 1-8-6.326V12.5a2.5 2.5 0 1 0 4 2V3h4.083A6.005 6.005 0 0 0 21 7.917\"/>",{"left":192,"top":192,"width":193,"height":193,"rotate":192,"vFlip":194,"hFlip":194,"body":219},"\u003Cpath fill=\"none\" stroke=\"currentColor\" stroke-linecap=\"round\" stroke-linejoin=\"round\" stroke-width=\"2\" d=\"M17 7L7 17M8 7h9v9\"/>",1782314630465]