Description
PrismML builds small, efficient AI models: 1-bit and ternary models called Bonsai, which the company says pack advanced reasoning, tool use and multimodal abilities into phones, laptops, wearables and efficient data centres. Its Bonsai 2 family spans 2 billion to 27 billion parameters, with memory footprints from about 0.43 GB to 5.9 GB. The company says the family uses far less memory and energy than standard models, and it publishes benchmark comparisons for intelligence density, throughput in tokens per second and energy per token across hardware platforms. The models are available to download, and there is a demo to try. The Bonsai Image model is also listed, along with a 16-bit standard version for comparison. The company points to a whitepaper for its evaluation method. PrismML is focused on AI research into efficiency, and it has a careers page for people who want to join the team.…
Highlights
- 1-bit and ternary Bonsai models
- Family from 2B to 27B parameters
- Models that run on phones and laptops
- Published benchmarks and energy numbers
- Downloadable models and a demo
Ratings & Reviews
No ratings yet
Be the first to rate PrismML.
People also visit
PrismML FAQ
What is PrismML?
PrismML builds small, efficient AI models for phones, laptops and data centres. Its Bonsai models use 1-bit and ternary weights.
What sizes are available?
The Bonsai 2 family runs from 2 billion to 27 billion parameters, with memory footprints from about 0.43 GB to 5.9 GB.
Can I try or download the models?
Yes. The site has download links for each model and a demo you can try.
How are the models evaluated?
The company publishes benchmark comparisons for intelligence density, throughput and energy use, and refers to a whitepaper for the method.
What are the best PrismML alternatives?
Popular alternatives to PrismML include Bonsai, Xybrid and Prismos AI. See all PrismML alternatives






