{"rewrite":{"id":"r_95cf31b269a3310594711383","clusterId":"c_898500c209c2cad535f52783","slug":"prismml-releases-ternary-bonsai-2-27b-at-a-5-9gb-footprint","model":"deepseek-v4-1-flash","headline":"PrismML Releases Ternary Bonsai 2 27B At A 5.9GB Footprint","summary":"PrismML announced Ternary Bonsai 2 27B, a compressed model built on Alibaba's Qwen3.8 27B. It ships at 5.9GB, one-ninth the size of the base model, and retains 98.2% of Qwen3.8 27B's benchmark scores. PrismML lists gains in reasoning, coding, vision, and long-horizon agent performance over the earlier Bonsai 27B. The model is released under the Apache 2.0 license and published on Hugging Face. PrismML says it runs on Mac, iPhone, and iPad.","whyItMatters":"PrismML is compressing a 27B model into a 5.9GB file that holds 98.2% of the base model's benchmark scores, which puts local execution within reach on graphics cards with only 8GB of VRAM.","webCardHtml":"\u003cp\u003eThe 5.9GB figure is what separates this release from the earlier Bonsai 27B. PrismML upgraded the base from Qwen3.6-27B to Qwen3.8 27B and reports the compressed result holds 98.2% of the full model\u0026#39;s benchmark scores, while the earlier release was marketed on running on an iPhone with heavy memory savings.\u003c/p\u003e\u003cp\u003eThe company lists reasoning, coding, vision, and long-horizon agent performance as improved over the previous generation. Weights are on Hugging Face under Apache 2.0, and PrismML names Mac, iPhone, and iPad as target hardware.\u003c/p\u003e","blueskyPost":"Ternary Bonsai 2 27B keeps 98.2% of Qwen3.8 27B's benchmarks at a ninth of the size, but PrismML ships it under Apache 2.0 on Hugging Face. The open license is the part that changes who can build on it.","twitterPost":"Ternary Bonsai 2 27B runs on Mac, iPhone, and iPad under Apache 2.0. The license, not the 5.9GB figure, decides who can ship with it.","threadsPost":"PrismML's Ternary Bonsai 2 27B compresses Qwen3.8 27B to 5.9GB and holds 98.2% of its benchmarks. The Apache 2.0 license and the Hugging Face release matter more than the size drop, because they set what other teams can do with the weights.","newsletterBlurb":"PrismML announced Ternary Bonsai 2 27B, a compressed build of Alibaba's Qwen3.8 27B that ships at 5.9GB, one-ninth the base model's size. The company reports it retains 98.2% of the base model's benchmark scores and improves on the earlier Bonsai 27B in reasoning, coding, vision, and long-horizon agent performance. Weights are on Hugging Face under Apache 2.0, and PrismML lists Mac, iPhone, and iPad.","attributionJson":"[{\"source\":\"GIGAZINE\",\"url\":\"https://gigazine.net/news/20260918-bonsai-2-27b/\",\"title\":\"AI model \\\"Ternary Bonsai 2 27B\\\" arrives, shrinking Qwen3.8 27B to 5.9GB while maintaining 98.2% performance\"}]","lintFlagsJson":null,"lintHits":0,"costUsd":0,"inputTokens":4855,"outputTokens":753,"status":"published","repairAttempts":0,"nextRepairAt":null,"factsAttemptedAt":1789698567,"createdAt":"2026-09-18T02:26:18.000Z","publishedAt":"2026-09-18T02:28:23.000Z","updatedAt":"2026-09-18T02:28:23.000Z"},"cluster":{"id":"c_898500c209c2cad535f52783","canonicalTitle":"Qwen3.8 27Bを5.9GBまで小型化しつつ性能は98.2％維持したAIモデル「Ternary Bonsai 2 27B」が登場","representativeArticleId":"a_0515cac1abf9d6fec8d73a25","sourceCount":1,"writtenSourceCount":1,"writeAttempts":0,"isSolo":true,"entitiesJson":"{\"anime_titles\":[],\"manga_titles\":[],\"work_titles\":[\"Ternary Bonsai 2 27B\",\"Qwen3.8 27B\"],\"studios\":[\"PrismML\",\"Alibaba\"],\"people\":[],\"type\":\"announcement\",\"domain\":\"other\",\"is_roundup\":false}","contentType":"news","status":"published","firstSeenAt":"2026-09-18T01:45:00.000Z","lastSeenAt":"2026-09-18T01:45:00.000Z","updatedAt":"2026-09-18T02:28:21.000Z"},"attribution":[{"source":"GIGAZINE","url":"https://gigazine.net/news/20260918-bonsai-2-27b/","title":"Qwen3.8 27Bを5.9GBまで小型化しつつ性能は98.2％維持したAIモデル「Ternary Bonsai 2 27B」が登場"}],"entities":{"anime_titles":[],"manga_titles":[],"work_titles":["Ternary Bonsai 2 27B","Qwen3.8 27B"],"studios":["PrismML","Alibaba"],"people":[],"type":"announcement","domain":"other","is_roundup":false},"keyFacts":null}