PrismML has developed and open-sourced the 1-bit Bonsai series of large language models, with the flagship 1-bit Bonsai 8B containing 8.2 billion parameters and occupying 1.15 GB of memory. The 1-bit Bonsai 8B is reported to be about 14 times more compressed than comparable 16-bit models while delivering comparable benchmark performance. PrismML highlights real-world inference speed—approximately 44 tokens per second on an iPhone 17 Pro Max—and says the models reduce energy consumption by 4 to 5 times compared to 16-bit equivalents. The company notes potential for further efficiency gains if future hardware supports 1-bit inference. PrismML was co-founded by Babak Hassibi and disclosed that its work was supported by a $16.25 million funding round led by Khosla Ventures, Cerberus Capital, and Caltech. Vinod Khosla is quoted in the coverage calling the model a major technological breakthrough.

Total raised

$16M

Funding rounds

1

Latest round

Equity

Latest activity

Apr 2026