PrismML's Bonsai 27B, based on Qwen3.6 27B, is a multimodal model with 262K-token context and tool-calling support, shipped in ternary (5.9 GB) and 1-bit (3.9 GB) variants.
The 1-bit build fits within an iPhone 17 Pro's memory budget for on-device inference; the ternary build targets laptop deployment. Both variants retain 90-95% of full-precision scores across 15 benchmarks and are released under Apache 2.0.
Ternary Bonsai 2 27B retains 98.2% of Qwen3.8 27B benchmark performance in a 5.9GB footprint, with multimodal and agentic capabilities.
