PrismML raises $22M seed to shrink AI models onto your phone
What's the deal? AI lab PrismML has raised $22.25 million in a seed round from Khosla Ventures, Cerberus CapitalDealroom has a profile for this one. Try Dealroom →, and CaltechDealroom has a profile for this one. Try Dealroom →. The startup, founded by a group of Caltech researchers, is building reasoning large language models small enough to run on PCs and smartphones.
Why now? On Thursday, PrismML released Bonsai 2 27B, which compresses Alibaba's widely used open-source Qwen3.8 27B model down to 5.9GB. That is a 9x to 10x reduction in memory versus the original, small enough to fit on a PC and possibly a high-end smartphone.
How it works: PrismML shrinks the "weights" that store what a model learns during training. Its approach, called "ternary" weights, cuts each weight from 16 bits down to three values: +1, −1, or 0.
Chief executive officer Babak Hassibi, a Caltech professor and compression expert, says the technique is unique because performance barely drops. Bonsai 2 matches 98% of Qwen's aggregate benchmark scores, up from 95% for the first Bonsai, released in March. That original model has been downloaded over 11 million times, with PrismML's smaller models adding another 2.6 million.
What's the endgame? PrismML wants to apply the technique to far bigger models. "The next models that we will release, hopefully in the next couple of months, will be in the several-hundred-billion-parameter range, and I expect it will be easier to retain the intelligence there," Hassibi said.
What could go wrong? Full benchmark parity may never arrive; compression will likely always have some impact, Hassibi says. PrismML is also not alone — Spain's Multiverse Computing works on similar tech and has raised more money.
The signal: At $22.25 million, PrismML's seed sits in the top 1% of all US enterprise software seed rounds on record. The backing signals investor appetite for a bet that capable AI need not be large — and could shift heavy computing from data centres onto everyday devices.
Read more: TechCrunch
Image credit: Generated with Gemini
