Apple is in discussions with Silicon Valley startup PrismML over technology that compresses large artificial intelligence models enough to run directly on iPhones.

This was disclosed by the company’s Chief Executive Officer, Babak Hassibi, according to CNBC.

On Tuesday, the Khosla Ventures-backed startup, which was spun out of the California Institute of Technology, United States unveiled compressed versions of Alibaba’s open-source Qwen AI model.

PrismML said it reduced the model’s size from about 54GB to under 4GB, enabling all 27 billion parameters to run on an iPhone 15 and newer devices.

Hassibi said that Apple, alongside several other technology companies, has been testing PrismML’s compressed AI models, evaluating their speed, energy efficiency and overall performance on mobile devices.

“They’re really evaluating our technology right now,” Hassibi said of Apple.

Hassibi described the discussions with Apple as being at a very early stage, adding that it remains uncertain how they will develop, although he said “things are progressing nicely.”

Apple did not immediately respond to CNBC’s request for comment.

The Information was the first to report PrismML’s AI model compression breakthrough.

The announcement comes a day after Apple released the public beta of iOS 27, giving iPhone users their first widespread access to the company’s long-awaited Siri upgrade.

Apple is seeking to make Siri more competitive with AI assistants developed by OpenAI and Anthropic, while ensuring that more personal data and artificial intelligence processing remain on users’ devices rather than being handled in the cloud.