Caltech mathematician Babak Hassibi's AI lab PrismML says it has run Alibaba's open-source Qwen3.6-27B model on an iPhone 17 Pro — after compressing it from roughly 54GB to under 4GB with no noticeable drop in performance.
Qwen3.6-27B is a dense 27-billion-parameter model, meaning all parameters are active during inference, unlike a mixture-of-experts model that activates only a subset.
PrismML claims the compressed version handles complex conversations, reasoning, agent tasks, and code generation. The company also says Apple has already reached out to discuss using the technology. Apple has long pushed to run more AI on-device, but advanced Siri features still rely on cloud models.