Siri, Apple’s oft-criticized voice assistant, is set to receive a 1.2 trillion parameter brain transplant, and the surgeon will be Google. Apple is reportedly on the verge of signing a $1 billion annual contract to license Google’s “ultrapowerful” Gemini AI model. This move is the core of a massive project to rebuild Siri’s underlying technology and deliver a new slate of intelligent features to users by next spring, finally addressing years of complaints about its limited capabilities.
This strategic, if somewhat humbling, partnership for Apple was finalized after an extensive evaluation of top AI models, including OpenAI’s ChatGPT and Anthropic’s Claude. Google’s Gemini was selected as the best-in-class “interim solution” to power the new Siri, code-named “Linwood.” The project, “Glenwood,” is a high-priority effort led by Apple’s software chief Craig Federighi and Vision Pro creator Mike Rockwell, aimed at closing the generative AI gap.
The 1.2 trillion parameter model from Google represents a seismic shift in capability for Siri. This massive model, which dwarfs Apple’s current 150-billion parameter cloud AI, will be specifically deployed to handle the assistant’s “summariser” and “planner” functions. This will allow Siri to understand complex, multi-part requests, synthesize information, and plan actions, functions that are currently the domain of more advanced AI chatbots.
Despite using Google’s core AI, Apple is maintaining its strict privacy stance. The Gemini model will be hosted entirely on Apple’s Private Cloud Compute servers. This “walled-off” architecture is a critical part of the deal, as it ensures all user data and interactions remain within Apple’s ecosystem and are never seen by Google. Apple is dedicating significant server resources to run this powerful model independently.
While the deal is a massive win for Google’s AI division, it will not be publicly advertised by Apple, which will treat Google as a silent technology supplier. This is a temporary measure, as Apple’s long-term goal is to replace Gemini with its own proprietary AI. An in-house 1 trillion parameter model is reportedly in development, but it faces the immense challenge of catching Google’s rapidly evolving, top-performing Gemini platform.