{"rewrite":{"id":"r_04e1661e414c3520c7937283","clusterId":"c_29115c5b785aabf401ec6709","slug":"linum-releases-p-jit-image-model-with-training-cost-kept-at-21-000","model":"deepseek-v4-1-flash","headline":"Linum Releases P-JiT Image Model With Training Cost Kept At $21,000","summary":"Linum, an AI research organization run by two AI researchers, has released the image generation model Pyramid-JiT (P-JiT) with open weights. P-JiT processes images directly as pixel data instead of converting them into latent representations, and its training stayed at roughly $21,000. The model reached convergence with about one-eleventh the samples of the earlier Linum v2 by compressing tokens harder.","whyItMatters":"P-JiT shows that a pixel-space image model can converge on about one-eleventh the samples and roughly $21,000 in training spend, which shifts the cost argument for researchers who cannot match large-lab budgets.","webCardHtml":"\u003cp\u003eP-JiT skips the usual VAE step. The model runs on 32x32 image patches rather than latent representations, which cuts the attention context window enough to train on far fewer samples than Linum v2. The architecture pairs a text encoder built from Qwen3.5-4B with a stack of Diffusion Transformer blocks, split into two single-modality refiner blocks and 22 identical blocks. Training ran in stages, with around 660 million samples at 512x512 before roughly 50 million at 1024x1024. Linum states the model does not handle text inside images, posters, or infographics.\u003c/p\u003e","blueskyPost":"Linum released P-JiT, a pixel-space image model trained for about $21,000. Harder token compression let it converge on roughly one-eleventh the samples of Linum v2. Open weights are out; text-in-image generation is not supported.","twitterPost":"Linum released P-JiT with open weights. The model works directly on pixel data, and harder token compression cut training to about one-eleventh the samples of Linum v2, at roughly $21,000 in cost.","threadsPost":null,"newsletterBlurb":"Linum has released Pyramid-JiT, an image generation model that works directly on pixel data instead of latent representations. The organization puts training cost at about $21,000, with convergence reached on roughly one-eleventh the samples of its earlier Linum v2. The model does not handle text inside images, posters, or infographics.","attributionJson":"[{\"source\":\"GIGAZINE\",\"url\":\"https://gigazine.net/news/20261109-pjit-21k/\",\"title\":\"P-JiT, an Image Generation AI Model That Achieved Efficient Training at Ultra-Low Cost, Now Available with Open Weights\"}]","lintFlagsJson":null,"lintHits":0,"costUsd":0,"inputTokens":5571,"outputTokens":707,"status":"published","repairAttempts":0,"nextRepairAt":null,"factsAttemptedAt":1794178107,"createdAt":"2026-11-08T22:42:39.000Z","publishedAt":"2026-11-08T22:47:25.000Z","updatedAt":"2026-11-08T22:47:25.000Z"},"cluster":{"id":"c_29115c5b785aabf401ec6709","canonicalTitle":"超低コストで効率的な学習に成功した画像生成AIモデル「P-JiT」、オープンウェイトで公開中","representativeArticleId":"a_a1c30ef24b374e0fb36b1418","sourceCount":1,"writtenSourceCount":1,"writeAttempts":0,"isSolo":true,"entitiesJson":"{\"anime_titles\":[],\"manga_titles\":[],\"work_titles\":[\"P-JiT\"],\"studios\":[\"Linum\"],\"people\":[],\"type\":\"news\",\"domain\":\"other\",\"is_roundup\":false}","contentType":"news","status":"published","firstSeenAt":"2026-11-08T22:00:00.000Z","lastSeenAt":"2026-11-08T22:00:00.000Z","updatedAt":"2026-11-08T22:47:26.000Z"},"attribution":[{"source":"GIGAZINE","url":"https://gigazine.net/news/20261109-pjit-21k/","title":"超低コストで効率的な学習に成功した画像生成AIモデル「P-JiT」、オープンウェイトで公開中"}],"entities":{"anime_titles":[],"manga_titles":[],"work_titles":["P-JiT"],"studios":["Linum"],"people":[],"type":"news","domain":"other","is_roundup":false},"keyFacts":["Linum released the image generation model Pyramid-JiT (P-JiT) with open weights.","P-JiT processes images directly as pixel data without converting them into latent representations.","P-JiT training cost approximately $21,000.","P-JiT converged with about one-eleventh the number of samples compared with Linum v2.","P-JiT does not support generation of images containing text, posters, or infographics."],"pageLinks":{"healthSlugs":["p-jit","linum"],"storySlugs":[],"entities":[{"slug":"p-jit","supported":0,"disputed":0,"contradicted":0,"unverified":2,"overall":"unverified","story_count":1,"has_dossier":false},{"slug":"linum","supported":0,"disputed":0,"contradicted":0,"unverified":1,"overall":"unverified","story_count":1,"has_dossier":false}],"stories":[],"entitiesOk":true,"storiesOk":true}}