PrismML — Concentrating intelligence
Concentrating intelligence
Large models can't fit on smartphones. Datacenters can't sustain them. PrismML is building ultra dense intelligence to solve both.
1-bit Bonsai→1-bit Bonsai Image→
14× less memory
8× faster
5× less energy
1-bit Bonsai
The first commercially viable models with 1-bit weights. Available in 8B, 4B, and 1.7B sizes, these models were engineered for robotics, real-time agents, and edge computing. They have a 14× smaller footprint than their full-precision counterparts, run 8× faster, and are 5× more energy efficient, while matching leading models at similar parameter counts on benchmarks. This results in over 10× the intelligence density of full-precision equivalents¹.
Ternary Bonsai
Ternary Bonsai models use {-1, 0, 1} weights to deliver a powerful balance between model quality and deployment efficiency. Available in 8B, 4B, and 1.7B sizes, these models have a 9× smaller footprint than full-precision counterparts and run roughly 5× faster, while delivering substantially stronger benchmark performance than most models at similar parameter counts. This creates a compelling tradeoff between capability and efficiency2.
Bonsai Image 4B
Available in 1-bit and Ternary variants, Bonsai Image 4B brings high-quality image generation to everyday devices. Built for local inference on iPhone, Mac, and GPUs, it reduces the diffusion transformer size by up to 8x and speeds image generation by up to 5.6× versus its full-precision counterparts, while preserving strong visual quality. The result is a lighter, and more deployable image generation model.
Download models↗Try Bonsai Image↗Bonsai Studio↗
Supported by:
Supported by:
.svg)
Benchmark palette
Intelligence density Bonsai 8B canopy Performance vs. size Benchmarks Throughput Energy consumption
Intelligence density
Negative log of the model's error rate divided by the model size
For detailed evaluation approach, please refer to our whitepaper.
Model benchmark comparison
Average score (IFEval, GSM8K, HumanEval+, BFCL, MuSR, MMLU-Redux)
Benchmark ← →
For detailed evaluation approach, please refer to our whitepaper.
Throughput
Tokens per second across hardware platforms (higher is better)
For detailed evaluation approach, please refer to our whitepaper.
Performance vs. size
Average score (IFEval, GSM8K, HumanEval+, BFCL, MuSR, MMLU-Redux)
For detailed evaluation approach, please refer to our whitepaper.
Bonsai 8B canopy
Average score (IFEval, GSM8K, HumanEval+, BFCL, MuSR, MMLU-Redux)
For detailed evaluation approach, please refer to our whitepaper.
Energy consumption
Milliwatt-hours per token across hardware (lower is better)
For detailed evaluation approach, please refer to our whitepaper.
16.0 GB 16-bit (standard)
1-bit Bonsai 8B
Centering AI research on efficiency
Successful artificial intelligence isn’t just about making models larger, but also smarter. Utilizing breakthrough research at Caltech, PrismML is pushing the frontier of intelligence density by reshaping how models are designed, prioritizing intelligence per bit over sheer parameter count.
Join the team
Select role...Staff AI/ML Engineer – Large-Scale SystemsStaff AI/ML Engineer – Edge & Consumer AISenior AI/ML Engineer – Kernel OptimizationSenior AI/ML Engineer – Post-Training PlatformAI/ML Engineer – Developer RelationsTechnical Recruiter – AI/ML Engineering
Resume:
Upload File
Max file size 10MB.
Uploading...
fileuploaded.jpg
Upload failed. Max size for files is 10 MB.
Thanks, your application has been received!
Oops! Something went wrong while submitting the form.
Contact us
Have a question, partnership idea, or a project that needs efficient intelligence? Reach out—we’d love to hear from you.
Thanks, your message has been received!
Oops! Something went wrong while submitting the form.