Ternary-weight (-1, 0, +1) compression of a 27B model by PrismML, shrunk to about 5.9 GB for local use.
Updated EU status checked Last shown
Sources: the vendor's own pages (linked above) and creator videos at the linked timestamps. EU status is information, not legal advice; see how we check.
Bonsai 2 27B is the largest model in PrismML's Ternary Bonsai 2 family, which spans 2B to 27B parameters. According to a YouTube review, it is a ternary-weight (-1, 0, +1) compression of Qwen 3.8 27B that reduces the model from about 55.6 GB to 5.9 GB, roughly nine times smaller, with benchmark performance reported as similar to the original. These figures come from a creator's summary and have not been independently verified here.
PrismML describes the Bonsai 2 family as bringing multimodal and agentic capabilities (reasoning, tool use) to local AI on personal computers, smartphones and wearables. Reported variants include an MLX build for Apple devices and Q1 (5.95 GB) and Q2 (7.21 GB) versions. PrismML's site presents benchmarks for intelligence density, throughput and energy consumption across hardware.
The provided material does not state the licence, download location, pricing or exact release date. The website lists other models in the family (9B, 4B, 2B) as well as a separate 1-bit Bonsai 8B and a Bonsai Image model.
Watch the moment: the exact second top creators show it. Videos load only when you press play, via youtube-nocookie.com.
Ternary-weight (-1, 0, +1) compression of Qwen 3.8 27B that shrinks it from 55.6 GB to 5.9 GB, over nine times smaller, with similar benchmark performance. Versions include MLX for Apple devices and Q1 (5.95 GB) and Q2 (7.21 GB).
Closest neighbours by meaning, not just by category.
Model · PrismML
Ternary/1-bit compressed 27B-parameter model from PrismML, about 3.9–5.9 GB, built to run on phones and laptops.
Model · PrismML
PrismML's 1-bit/ternary compressed Flux 2 Klein text-to-image model (~1 GB) for offline on-phone generation.
Model · Alibaba
Compact 27B-parameter model from Alibaba's Qwen family, runnable locally on a single high-end consumer GPU.
Model · Alibaba
Small 0.8B-parameter open-weights language model in Alibaba's Qwen 3.5 family, aimed at lightweight and on-device use.
Model · Alibaba
Alibaba's Qwen 3.5 large open-weights language model (~122B parameters) from the Qwen family.
Model · Alibaba
Small 0.5B-parameter open-weights base language model from Alibaba's Qwen2.5 series, Apache 2.0 licensed.
