Prism ML develops ultra‑dense AI models that can run on edge devices, delivering large‑scale reasoning capabilities with dramatically reduced memory, speed, and power requirements. Its Bonsai 27B model, available in 1‑bit and ternary formats, fits on a smartphone (as small as 3.9 GB) while providing multi‑step reasoning, tool calling, and multimodal understanding, enabling sophisticated AI workflows locally without reliance on datacenters.
Funding
Funding not disclosed
Founders
Product
Problem
Large language models require substantial memory, compute, and energy, making them unsuitable for deployment on smartphones, laptops, or other edge devices. This limits on‑device AI capabilities, forces reliance on cloud services, and raises latency, privacy, and cost concerns.
Solution
Prism ML builds ultra‑dense AI models that compress model size while preserving multi‑step reasoning, tool calling, and multimodal understanding. Their Bonsai 27B model is offered in 1‑bit (3.9 GB) and ternary (5.9 GB) formats, delivering 14× lower memory usage, 8× faster inference, and 5× reduced energy consumption compared with conventional 27B models. The compact footprint enables the model to run locally on smartphones and laptops without external servers. By keeping inference on‑device, Prism ML provides low‑latency responses, eliminates data transmission, and cuts operational costs for edge applications.
Target Audience
Primary customers are developers and product teams building AI‑enabled applications for mobile phones, laptops, robotics, and Internet‑of‑Things devices that require on‑device intelligence.
Features
- 27‑billion‑parameter Bonsai model available in 1‑bit (3.9 GB) and ternary (5.9 GB) quantizations
- 14× reduction in memory footprint versus standard 27B models
- 8× faster inference speed on typical mobile and laptop hardware
- 5× lower energy consumption during inference
- Supports multi‑step reasoning, tool calling, agentic workflows, and multimodal inputs
- Designed for on‑device execution on smartphones, laptops, robots, and IoT devices