Apple in talks with startup firm to shrink AI models to run directly on iPhone
Apple in talks with startup firm to shrink AI models to run directly on iPhone
Publish Date: 2026-07-15 03:39:00
Source Domain: www.thestandard.com.hk
Apple is reportedly in talks with Silicon Valley startup PrismML to bring high-performance large AI models to the iPhone through model compression.
PrismML chief executive Babak Hassibi told CNBC that the company can shrink AI models to a size sufficient to run directly on the iPhone. Apple has been evaluating the company’s technology and measuring its speed, energy efficiency, and performance on devices.
PrismML released a compressed version of Alibaba’s (9988) open-source Qwen model on Tuesday, reducing the model from 54GB to 4GB, achieving a memory usage reduction of over 90 percent and allowing all 27 billion parameters to run on an iPhone 15 or newer models.
Hassibi acknowledged a trade-off, as efficiency gains from compression come at the cost of model accuracy. The PrismML models typically lose several percentage points of overall performance, with factual recall weakening before reasoning, mathematical, and programming abilities.
Apple has launched a new version of Siri on Monday, in bid to compete with companies like OpenAI and Anthropic, while keeping more personal information and AI processing on-device.
While Apple can send complex requests to cloud-based models, running AI directly on the iPhone would reduce the delay associated, lower cloud-computing costs and support the company’s privacy pitch.
