{"apiVersion":"v1","taxonomyVersion":1,"generatedAt":"2026-10-29T14:07:00.034Z","data":{"recordId":"wsa_793vjjqyuh54","score":0.864,"confidence":"high","capped":false,"capReason":null,"sources":[{"n":1,"url":"https://prismml.com/news/bonsai-2-27b","publisher":"PrismML","authority":"owner-controlled","verification":"verified","lastRetrievedAt":"2026-10-27T01:17:35.750Z","detectedPublishedAt":"2026-09-17T00:00:00.000Z","rights":"original"},{"n":2,"url":"https://prismml.com/news/prismml-launches-bonsai-2-27b","publisher":"PrismML","authority":"owner-controlled","verification":"verified","lastRetrievedAt":"2026-10-27T01:17:35.750Z","detectedPublishedAt":"2026-09-17T00:00:00.000Z","rights":"original"},{"n":3,"url":"https://huggingface.co/prism-ml/Ternary-Bonsai-2-27B-gguf-dev","publisher":"PrismML","authority":"owner-controlled","verification":"verified","lastRetrievedAt":"2026-10-27T01:17:35.750Z","detectedPublishedAt":"2026-09-17T08:31:00.000Z","rights":"original"},{"n":4,"url":"https://huggingface.co/prism-ml/Ternary-Bonsai-2-27B-gguf","publisher":"Hugging Face","authority":"other","verification":"verified","lastRetrievedAt":"2026-10-27T01:17:35.750Z","detectedPublishedAt":"2026-09-16T23:40:56.000Z","rights":"original"},{"n":5,"url":"https://huggingface.co/prism-ml/Ternary-Bonsai-2-27B-mlx-2bit","publisher":"Hugging Face","authority":"other","verification":"verified","lastRetrievedAt":"2026-09-24T04:13:10.866Z","detectedPublishedAt":"2026-09-16T23:41:18.000Z","rights":"original"}],"claims":[{"id":"591016ee-6840-4c8a-8670-d9dea0500117","kind":"capability","statement":"Bonsai 2 needs an activation transform that currently only exists in PrismML’s llama.cpp fork.","citations":[{"source":3,"supportProbability":0.91,"supports":true,"quote":{"text":"Bonsai 2 needs an activation transform that currently only exists in our","rights":"original"}},{"source":4,"supportProbability":0.81,"supports":true,"quote":null}]},{"id":"0c53b9d4-2d60-4875-9539-a821766b5a97","kind":"availability","statement":"The Q2_0 packing is kept separate until upstream support lands.","citations":[{"source":3,"supportProbability":0.91,"supports":true,"quote":{"text":"The `Q2_0` packing of Bonsai 2 27B, kept separate until upstream support lands.","rights":"original"}}]},{"id":"64b91f78-27e4-417a-ace6-b69ca083548a","kind":"availability","statement":"The build will move to the main repo once upstream support lands.","citations":[{"source":3,"supportProbability":0.57,"supports":false,"quote":{"text":"It moves to the main repo once that lands.","rights":"original"}}]},{"id":"b15e9859-abf6-4aa9-b49a-aa7bb76bf93e","kind":"recommendation-relevant","statement":"The build is intended for testing, kernel work, and upstreaming.","citations":[{"source":3,"supportProbability":0.69,"supports":false,"quote":{"text":"It is here for testing, kernel work, and upstreaming.","rights":"original"}}]},{"id":"e9fdd733-3511-4ffd-be03-f6ec534c55c2","kind":"capability","statement":"Stock llama.cpp may load the file and output gibberish with no warning.","citations":[{"source":3,"supportProbability":0.63,"supports":false,"quote":null},{"source":4,"supportProbability":0.81,"supports":true,"quote":null}]},{"id":"a9b830b6-1b7d-4215-ad07-e6c381855456","kind":"capability","statement":"The model supports multimodal text-and-image input and a 262K-token context window.","citations":[{"source":1,"supportProbability":0.98,"supports":true,"quote":{"text":"It supports a 262K-token context window , multimodal text-and-image input, and is released under the Apache 2.0 license.","rights":"original"}}]},{"id":"3d3b9a77-d8ef-46b6-9a93-e3d3c64917ae","kind":"capability","statement":"The model uses ternary weights with FP16 group-wise scaling and has a total model footprint of 5.9GB.","citations":[{"source":1,"supportProbability":0.98,"supports":true,"quote":{"text":"Ternary Bonsai 2 27B uses ternary {−1, 0, +1} weights with FP16 group-wise scaling, for 1.76 effective bits per weight and a total model footprint of 5.9GB .","rights":"original"}}]},{"id":"3e7f79d5-0722-4bbc-b57d-ce02dad7b978","kind":"deployment","statement":"The model runs on NVIDIA GPUs via CUDA and Apple devices via MLX.","citations":[{"source":1,"supportProbability":0.98,"supports":true,"quote":{"text":"Bonsai 2 27B runs on NVIDIA GPUs via CUDA and on Apple devices (Mac, iPhone, iPad) via MLX, through custom low-bit kernels.","rights":"original"}}]},{"id":"c7e210b8-fe67-48a1-9183-42e8c3b9eed7","kind":"recommendation-relevant","statement":"The model is intended for local assistants, multimodal agents, private knowledge workflows, computer-use applications, and long-running agentic tasks.","citations":[{"source":2,"supportProbability":0.98,"supports":true,"quote":{"text":"It is intended for local assistants, multimodal agents, private knowledge workflows, computer-use applications, and long-running agentic tasks that have traditionally depended heavily on cloud inference.","rights":"original"}}]},{"id":"ffaefdc6-8809-4151-910a-c0ce8ec43400","kind":"availability","statement":"Users can download the model for free starting September 17, 2026.","citations":[{"source":2,"supportProbability":0.97,"supports":true,"quote":{"text":"Developers, researchers, and other users can download the Ternary Bonsai 2 27B models under the Apache 2.0 license for free starting September 17, 2026.","rights":"original"}}]},{"id":"cc1fbf0d-6e38-492d-86dc-025849aac5ed","kind":"deployment","statement":"The ternary model is 5.9 GB and reduces memory footprint by more than 9x compared with its full-precision counterpart.","citations":[{"source":2,"supportProbability":0.98,"supports":true,"quote":{"text":"At just 5.9 GB, Ternary Bonsai 2 27B reduces memory footprint by more than 9x compared with its full-precision counterpart.","rights":"original"}}]},{"id":"45c4406d-f6bc-450c-9c73-1c6fbc616af2","kind":"capability","statement":"It retained 98.2% of Qwen3.8 27B’s benchmark performance.","citations":[{"source":2,"supportProbability":0.96,"supports":true,"quote":{"text":"However, while the original Ternary Bonsai 27B retained 95% of the aggregate benchmark performance of its full precision counterpart, the new Ternary Bonsai 2 27B retained over 98%.","rights":"original"}}]},{"id":"75a81c61-92da-40e5-a657-b4955317e706","kind":"capability","statement":"It achieved an aggregate score of 83.9 across a 20-benchmark suite.","citations":[{"source":2,"supportProbability":0.98,"supports":true,"quote":{"text":"Across a 20-benchmark suite spanning reasoning, math, coding, instruction following, vision, and agentic tool use, Ternary Bonsai 2 27B achieves an aggregate score of 83.9 , retaining 98.2% of Qwen3.8 27B’s performance of 85.4 .","rights":"original"}}]},{"id":"217cd42d-15d4-4947-a1cf-f933d0237227","kind":"capability","statement":"The model is a 27.8-billion-parameter multimodal language model.","citations":[{"source":2,"supportProbability":0.99,"supports":true,"quote":{"text":"The Ternary Bonsai 2 27B model is a 27.8-billion parameter multimodal language model.","rights":"original"}}]}]},"nextCursor":null,"license":{"catalog":{"name":"CC BY 4.0","spdx":"CC-BY-4.0","url":"https://creativecommons.org/licenses/by/4.0/","attribution":"wowsuchai","attributionUrl":"https://wowsuchai.withglint.app","covers":["catalog metadata","classifications","generated summaries","feed event metadata"]},"excluded":{"covers":["source quotations","linked source content"],"terms":"Not licensed by wowsuchai. Source quotations and linked source content keep their original rights."}}}