14:00 ETPrismML Introduces Ternary Bonsai Model Family

Delivering top intelligence at just 1.58 bits per weight , /PRNewswire/ -- PrismML, a pioneer in high-performance AI models, today announced the Ternary Bonsai model family: three state-of-the-art large language models available in 8B, 4B, and 1.7B parameter sizes, built on a novel 1.58-bit weight architecture. Building on their recent introduction of the world's first commercially viable 1-bit models, PrismML continues to push the boundaries of efficient AI. From laptops and edge devices to embedded systems and large-scale datacenter infrastructure, the company is advancing a new paradigm centered on intelligence density - the amount of reasoning capability delivered per unit of memory, compute, and energy. "Intelligence density is a defining metric of next-generation AI," said Babak Hassibi, CEO and Founder of PrismML and Professor at Caltech. "With Ternary Bonsai, we deliver significantly more reasoning capability while dramatically reducing memory, energy, and hardware requirements. This unlocks powerful AI in environments where it was previously impractical or impossible." About Ternary Bonsai model family:Ternary Bonsai models use a ternary weight representation, where each weight takes one of three values: {-1, 0, +1}, corresponding to 1.58 bits per weight. PrismML's proprietary architecture applies this representation end-to-end across the entire network. At just 1.58 bits per weight, the models achieve dramatically smaller memory footprints compared to standard 16-bit models: Ternary Bonsai 8B: ~1.75 GB (vs. ~16.4 GB) Ternary Bonsai 4B: ~0.86 GB (vs. ~8 GB) Ternary Bonsai 1.7B: ~0.37 GB (vs. ~3.4 GB) This represents roughly a 9× reduction in memory usage across all model sizes. Compared to PrismML's 1-bit Bonsai 8B model, the Ternary Bonsai 8B delivers an average improvement of 5 benchmark points while requiring only ~600 MB of additional memory. Despite their compact size, all three models outperform most leading full-precision models within their respective parameter classes on standard benchmarks. Ternary Bonsai models are fully ternary end-to-end. There are no higher-precision escape hatches. Embeddings, attention layers, MLPs, and the LM head all use the same 1.58-bit representation. The models employ a group-wise quantization scheme, where each weight is constrained to one of three values: {-s, 0, +s}. These three states are encoded as (-1, 0, +1) using 1.58 bits per weight, along with a shared FP16 scale factor (s) for each group of 128 weights. From Edge to Datacenter: Intelligence Density at Every Scale While the immediate impact of the Ternary Bonsai family is enabling powerful on-device AI with minimal hardware requirements, the benefits extend well beyond the edge. The same efficiency gains allow datacenters to improve hardware utilization, lower operating costs, and significantly reduce energy consumption. As the cost of AI infrastructure continues to rise, intelligence density is emerging not only as a product differentiator, but as a strategic priority for the industry. Pricing and Availability:Developers, researchers, and other users can download the Ternary Bonsai family of models under the Apache 2.0 license for free starting today. Download 1-bit Ternary Bonsai Models here About PrismML:PrismML is a U.S.-based artificial intelligence company focused on making AI more efficient and accessible. PrismML is built on proprietary Caltech intellectual property and backed by Khosla Ventures, Cerberus Ventures, and compute grants from Google and Caltech. For more information, visit the Website, LinkedIn, or X. All registered trademarks and product identifiers belong to their respective corporate entities. Any other trademarks or product names referenced here are also owned exclusively by their relevant companies. Media ContactGary BirdPrismML[email protected]831.888.9011 SOURCE PrismML