{"rewrite":{"id":"r_00c648d522580653512f7040","clusterId":"c_3bf2476cf5aa8d1827796da1","slug":"smolmaxxing-runs-a-16-billion-parameter-model-on-a-1gb-ipad-air","model":"deepseek-v4-1-flash","headline":"smolmaxxing Runs a 16-Billion-Parameter Model on a 1GB iPad Air","summary":"An open-source project called smolmaxxing modifies llama.cpp so large AI models run on devices with very little memory. On a Xiaomi TV Box S with 1.9GB, the 16-billion-parameter Ling-mini-2.0 ran at 5.1 tokens per second. On a first-generation iPad Air with 1GB, Ling-mini-2.0 ran at 4.5 tokens per second and the 1.7-billion-parameter Bonsai-1.7B at 23.7 tokens per second. The iPad Air test required jailbreaking.","whyItMatters":"The project shows local AI performance is now limited as much by model structure and memory handling as by raw device memory, since a 9-billion-parameter model on the same TV box managed only 0.27 tokens per second.","webCardHtml":"\u003cp\u003eThe developer ran about 160 experiments and published both the program and the results on GitHub, with optimized models hosted on Hugging Face. The memory approach keeps frequently used data in memory and reads the rest from storage as needed, with processing tuned for the CPU.\u003c/p\u003e\u003cp\u003eSpeed depends heavily on model structure. On the Xiaomi TV Box S, the 9-billion-parameter Qwen3.5-9B managed 0.27 tokens per second, and the developer names Qwen3.5-2B as the model with the best balance of speed and quality. The stated aim is to reach smartphones and routers next.\u003c/p\u003e","blueskyPost":"smolmaxxing hit 4.5 tokens per second on a 1GB iPad Air only after a jailbreak, so the memory ceiling it clears is real but the install path is not one most owners can follow.","twitterPost":"smolmaxxing's 1GB iPad Air run needed a jailbreak, so the memory limit it clears is not the same as an install path.","threadsPost":"smolmaxxing ran Ling-mini-2.0 on a 1GB iPad Air at 4.5 tokens per second, but that test required jailbreaking the device. The memory ceiling it clears is real. The install path is the part most owners cannot follow.","newsletterBlurb":"smolmaxxing is an open-source modification of llama.cpp that runs large models on devices with very little memory. The developer reports Ling-mini-2.0 at 5.1 tokens per second on a 1.9GB Xiaomi TV Box S and 4.5 tokens per second on a jailbroken first-generation iPad Air with 1GB. About 160 experiments are documented on GitHub.","attributionJson":"[{\"source\":\"GIGAZINE\",\"url\":\"https://gigazine.net/news/20261113-smolmaxxing/\",\"title\":\"Free tool \\\"smolmaxxing\\\" lets you run a 16-billion-parameter AI locally even on old iPads and small devices with 1GB of memory\"}]","lintFlagsJson":null,"lintHits":0,"costUsd":0,"inputTokens":5658,"outputTokens":868,"status":"published","repairAttempts":0,"nextRepairAt":null,"factsAttemptedAt":1794573877,"createdAt":"2026-11-13T12:39:57.000Z","publishedAt":"2026-11-13T12:43:24.000Z","updatedAt":"2026-11-13T12:43:24.000Z"},"cluster":{"id":"c_3bf2476cf5aa8d1827796da1","canonicalTitle":"古いiPadやメモリ1GBの小型端末でも160億パラメータのAIをローカルで動かせる無料ツール「smolmaxxing」が登場","representativeArticleId":"a_cf665cc5b9b6a641d381ebcb","sourceCount":1,"writtenSourceCount":1,"writeAttempts":0,"isSolo":true,"entitiesJson":"{\"anime_titles\":[],\"manga_titles\":[],\"work_titles\":[\"smolmaxxing\"],\"studios\":[],\"people\":[],\"type\":\"news\",\"domain\":\"other\",\"is_roundup\":false}","contentType":"news","status":"published","firstSeenAt":"2026-11-13T12:00:00.000Z","lastSeenAt":"2026-11-13T12:00:00.000Z","updatedAt":"2026-11-13T12:43:22.000Z"},"attribution":[{"source":"GIGAZINE","url":"https://gigazine.net/news/20261113-smolmaxxing/","title":"古いiPadやメモリ1GBの小型端末でも160億パラメータのAIをローカルで動かせる無料ツール「smolmaxxing」が登場"}],"entities":{"anime_titles":[],"manga_titles":[],"work_titles":["smolmaxxing"],"studios":[],"people":[],"type":"news","domain":"other","is_roundup":false},"keyFacts":["smolmaxxing is an open-source project that modifies llama.cpp to run AI models on devices with limited memory.","The 16-billion-parameter Ling-mini-2.0 ran at 5.1 tokens per second on a Xiaomi TV Box S with 1.9GB of memory.","Ling-mini-2.0 ran at 4.5 tokens per second on a first-generation iPad Air with 1GB of memory, which required jailbreaking.","The 1.7-billion-parameter Bonsai-1.7B ran at 23.7 tokens per second on the first-generation iPad Air.","The developer ran about 160 experiments and published the program and results on GitHub, with models on Hugging Face."],"pageLinks":{"healthSlugs":["smolmaxxing"],"storySlugs":[],"entities":[{"slug":"smolmaxxing","supported":0,"disputed":0,"contradicted":0,"unverified":0,"overall":"unverified","story_count":1,"has_dossier":false}],"stories":[],"entitiesOk":true,"storiesOk":true}}