PrismML Adds Smaller Models for On-Device AI
PrismML expands its Bonsai 2 family with three new, compact AI models designed for efficient operation on personal devices. The models offer enhanced reasoning and tool-calling capabilities.
Pasadena, Calif. – October 15, 2026 – AI company PrismML has released three new models for its Bonsai 2 family: 2B, 4B, and 9B parameter versions.
The models, based on the Qwen3.5 architecture, are optimized for operation on resource-constrained devices such as personal computers, phones, and wearables. PrismML's proprietary ternary weight representation enables smaller model footprints and improved energy efficiency, with minimal performance degradation compared to full-precision models.
These new releases enhance reasoning capabilities, visual understanding, and tool-calling functions. The company states that the models retain over 90% of the performance of full-precision Qwen3.5 models on tasks related to coding, math, reasoning, and tool invocation.
Babak Hassibi, Founder and CEO of PrismML, explained the goal is to provide developers with more capability within the constraints of on-device AI. The models demonstrate significant improvements in energy efficiency and throughput compared to previous generation Bonsai models.
The Bonsai 2 models are available for free download under the Apache 2.0 license. PrismML previously demonstrated local AI smart glasses operation on a Snapdragon platform, and these new models are intended to support similar applications for future wearable devices.