[News] PrismML Reportedly Shrinks Alibaba’s Qwen 3.6 to Run on iPhone 17 Pro, Drawing Apple’s Interest

Apple is continuing to strengthen its on-device AI strategy, and one startup has reportedly drawn the company’s interest. According to Seeking Alpha, citing The Information, PrismML has compressed Alibaba’s open-source Qwen 3.6 large language model to run on the iPhone 17 Pro. The reported breakthrough has attracted Apple’s attention as the company looks to expand its on-device AI capabilities. Apple has reportedly met with PrismML to discuss potential applications of the technology.

The milestone reflects Apple’s broader push to run AI directly on devices rather than relying on cloud infrastructure. As noted by MacRumors, running larger AI models directly on iPhones would allow more Apple Intelligence features to operate on-device instead of through Apple’s Private Cloud Compute servers. This could help reduce Apple’s operating costs while further enhancing user privacy.

PrismML Shrinks Large AI Models for Mobile Devices

As The Information notes, PrismML uses a mathematical technique to compress the Qwen 3.6 model to a fraction of its original size. While model compression typically comes at the expense of performance, the company says its approach reduces model size without degrading AI capabilities. PrismML has reportedly reduced Qwen 3.6 from approximately 54 GB to less than 4 GB. PrismML says its compressed model running on the iPhone is capable of handling complex conversations, reasoning, fully autonomous agents, and software coding. According to the report, the open-source model is scheduled for release on July 14.

The achievement is particularly notable because the largest AI models, which contain trillions of parameters, remain too large to run directly on mobile devices. As noted by STAR Market Daily, parameter count is commonly used as a measure of an AI model’s capability, yet most smartphones and laptops today can generally run only models with billions of parameters due to limited processing power and memory capacity.

Another notable feature of PrismML’s model is that it keeps all 27 billion parameters active simultaneously. The Information notes that one of Apple’s new on-device AI models has 20 billion parameters but uses a sparse architecture, with only 1 billion to 4 billion parameters active at any given time.

The Company Behind the Technology

PrismML is a spinout from the California Institute of Technology (Caltech), led by CEO Babak Hassibi, who is also a professor of electrical engineering at the university. Hassibi and his co-founders developed the mathematical techniques that underpin the startup’s technology. Caltech owns the underlying patents and has granted PrismML exclusive licensing rights. Seeking Alpha adds that Khosla Ventures, OpenAI’s first venture capital investor, participated in PrismML’s US$16.25 million seed funding round earlier this year.

Read more

• [News] Apple’s First Foldable iPhone Nears Fall Debut: TSMC 2nm A20 Pro, Crease-Free Display Lead Rumored Features
• [News] China AI Model Developers Ramp Up In-house Chips: DeepSeek, Zhipu Explore Custom Silicon

(Photo credit: FREEPIK)