Apple is always looking for ways to run advanced artificial intelligence directly on your device rather than relying on remote servers. The core challenge has always been the massive size of the data required to make these modern features work well. Now, a new startup called PrismML has found a practical way to shrink huge models so they can fit on a phone, and the tech giant is paying close attention to its progress.
PrismML shrinks massive 54GB language models down to just 4GB
The startup recently showed its ability to take the 54GB Qwen 3.6 model and compress it to only 4GB. This is a massive shift because a model of that size usually packs 27 billion parameters. Having that many parameters gives it the power to handle complex chat reasoning and even operate fully autonomous agents. Most models built for an iPhone only hold a few billion parameters, which severely limits what they can actually do.
What makes PrismML stand out is that its math-based compression trick does not hurt the model’s performance. The startup managed to fit this highly capable system onto a smartphone while keeping the fast response times that users expect from their devices.
Apple holds meetings to bring this compressed technology to smartphones
Because the company strongly prefers to process data right on the device for privacy reasons, finding a reliable way to compress large AI models is a major priority. A recent report from The Information revealed that representatives from the hardware maker have already held meetings with PrismML to discuss potential partnerships.
There is no official deal in place right now. Even if the two companies agree to work together, it remains unclear exactly when device owners or Siri users will see these upgrades. Still, the fact that these discussions are happening shows a clear push to make mobile devices much smarter without leaning on the cloud.
Moving heavy processing away from remote data centers and directly onto the phone hardware gives users more privacy and faster answers. If PrismML can deliver on its promises at a wider scale, the next generation of mobile software could finally rival the smartest cloud tools available right now.
