PrismML

PrismML is an artificial intelligence company developing highly efficient AI models designed to run on a wide range of devices, from smartphones to data centers. The company emerged from stealth on March 31, 2026, introducing what it describes as the world's first commercially viable 1-bit large language models (LLMs). This technology is based on proprietary intellectual property and research from the California Institute of Technology (Caltech).

The founding team includes CEO Babak Hassibi, a professor at Caltech, alongside Sahin Lale (Co-Head of Research), Omead Pooladzandi (Co-Head of Research), and Reza Sadri (VP, Strategy). Their work addresses the challenge of deploying large, computationally intensive AI models by fundamentally redesigning neural networks. Instead of using traditional 16- or 32-bit architectures, PrismML creates models with a native 1-bit structure, which significantly reduces the computational and memory requirements for inference without sacrificing performance.

PrismML's flagship product is the Bonsai family of models, which includes 8B, 4B, and 1.7B parameter versions. The 1-bit Bonsai 8B model, for example, has a memory footprint of just 1.15 GB, making it 14 times smaller and 8 times faster than comparable full-precision models. This allows it to run efficiently on consumer devices like iPhones and laptops, enabling real-time, on-device AI applications that are not limited by latency, hardware, or privacy concerns. The models are available for developers and researchers to download under the Apache 2.0 license.

The company's business model is focused on enabling a future where powerful AI can run locally and securely, while also allowing data centers to operate with greater energy and cost efficiency. By concentrating on "intelligence density"—the useful intelligence a model delivers per unit of size and power—PrismML aims to shift the paradigm from simply building larger models to creating smarter, more accessible ones. The company is backed by investors including Khosla Ventures, Cerberus Ventures, and has received compute grants from Google and Caltech.

Keywords: 1-bit AI models, large language models, edge AI, on-device AI, intelligence density, Caltech, neural network compression, Bonsai models, efficient AI, local AI deployment, high-performance AI, low-latency inference, AI for edge devices, machine learning, artificial intelligence, Khosla Ventures, Cerberus Ventures, AI hardware optimization, energy-efficient AI