PrismML compresses 27B parameter Qwen model to run on iPhone Apple interest
On-device AI just took a jump: startup PrismML says it compressed Alibaba's $BABA 27-billion-parameter Qwen model from 54 GB to under 4 GB and ran it on an iPhone 17 Pro, claiming no benchmark loss. Apple $AAPL has reportedly reached out. Per Seeking Alpha and Investing.com.
A small startup may have moved the needle on one of the biggest constraints in consumer AI: model size. PrismML, which raised a $16.25 million seed round earlier this year with participation from Khosla Ventures, says it has compressed Alibaba's $BABA open-source 27-billion-parameter Qwen model to run locally on an iPhone 17 Pro, according to reports from Seeking Alpha and Investing.com. The model was reportedly shrunk from 54 GB to under 4 GB while keeping all 27 billion parameters active, with the company claiming no loss of benchmark performance. That would be the largest AI model ever run on an iPhone, and the claim has apparently gotten attention where it matters: Apple $AAPL has reportedly reached out to the company. Apple has been pushing to run more AI directly on its devices rather than in the cloud, both for privacy positioning and to reduce dependence on external compute. The caveats are real. These are company claims, benchmark results have not been independently verified, and compression techniques often show their weaknesses only in extended real-world use. PrismML says it will release the compressed model as open source, which would let outside researchers test the claims directly. If they hold, the economics of on-device AI change: models that today require a data center connection could run in your pocket. Per Seeking Alpha and Investing.com.