Saturday, October 17, 2026 Privacy check Newsletter Advertise
Latest
AI news

PrismML Expands Bonsai 2 Family With New On-Device AI Models

PrismML has launched 2B, 4B, and 9B ternary models designed to bring advanced reasoning and tool-calling to on-device AI applications.

PrismML Expands Bonsai 2 Family With New On-Device AI Models. Source: Company press releases (PR Newswire)

PrismML expanded its Bonsai 2 family on October 15, 2026, introducing new 2B, 4B, and 9B ternary models designed for resource-constrained edge devices and personal computers.

Model Specifications and Footprints

PrismML announced the expansion of its Ternary Bonsai 2 family with the addition of 2B, 4B, and 9B models based on Qwen3.5. The models utilize the company's ternary weight representation, which encodes model weights using three values: minus one, zero, and plus one.

The new releases feature compact footprints of 0.43GB, 0.95GB, and 2.00GB respectively, making them significantly smaller than their full-precision counterparts. The expanded Ternary Bonsai 2 family now spans four sizes ranging from 2B to 27B parameters.

Performance and Efficiency

According to PrismML, the new models match or exceed full-precision counterparts across a suite of 19 benchmarks covering coding, math, and instruction following tasks. They retain over 90 percent of the benchmark scores of full-precision Qwen3.5 models on vision, agentic work, tool-calling, knowledge, and reasoning tasks.

On hardware such as the RTX 4090 and Macbook M5 Pro, the models deliver up to 3x the energy efficiency and 5x the throughput of full-precision models. On mobile deployments like an iPhone 18 Pro Max, the 9B model reaches up to 34 tokens per second, while the 2B model reaches up to 120 tokens per second.

Availability and Training

The Ternary Bonsai 2 2B, 4B, and 9B models were trained using Google v5 TPUs and are optimized for low-latency inference on consumer-grade CPUs and edge GPUs. PrismML stated that developers and researchers can download the models for free under the Apache 2.0 license starting October 15, 2026.

Key facts and where they come from
  • PrismML expanded its Bonsai 2 family with new 2B, 4B, and 9B ternary models based on Qwen3.5.
    PrismML today expanded its Bonsai 2 family with new 2B, 4B, and 9B ternary models based on Qwen3.5.
  • The models encode model weights using three specific values: -1, 0, and +1.
    encodes model weights using three values: {−1, 0, +1}.
  • The models are available for free download under the Apache 2.0 license starting October 15, 2026.
    download the new Bonsai 2 models under the Apache 2.0 license for free starting October 15, 2026.

Read the original from Company press releases (PR Newswire) →

Add TechUpscale as a preferred source on Google

The TechUpscale Brief

The day's cyber, AI and tech news in one short email, every weekday morning. Free. Unsubscribe anytime.

I'm most interested in

More AI

YYForce Launches Yolara AI Workforce Platform in Singapore

United StatesCompany press releases (PR Newswire)

CPG Radar Launches AI Market Intelligence and Sales Tools

United StatesCompany press releases (PR Newswire)