PrismML's Bonsai 2 27B hits 98% of Qwen at 5.9 GB

PrismML was founded by a group of Caltech researchers and is led by Babak Hassibi, a Caltech professor and compression expert. It has raised a $22.25 million seed round from Khosla Ventures, Cerberus Capital, and Caltech, and counts Ion Stoica, a Databricks co-founder and director of Berkeley's Sky Computing Lab, as an adviser.

This is not the only effort in LLM compression — Multiverse Computing, founded by a professor from Spain's Donostia International Physics Center, works on it too and has raised gobs of cash. Hassibi argues PrismML's tech stands out because its models have lost virtually no performance compared with the originals, and the benchmark match has improved from 95% to 98% across two releases. The original Bonsai has already been downloaded over 11 million times, and PrismML's even smaller models another 2.6 million times, the company says.

The next test is scale. Hassibi expects to release models in the several-hundred-billion-parameter range in the next couple of months, and predicts it will be easier to retain intelligence as models get bigger. Whether the compression holds up at that size — and whether it translates into everyday device use — is the open question.