Siri, which somehow still stinks, is finally getting a new core AI model, and it’s going to be a 1.2 trillion parameter Google Gemini-based LLM, according to Bloomberg Apple reporter Mark Gurman. Until Apple gets its act together on a proprietary AI system, Gurman writes, it will reportedly give Google about $1 billion per year for a temporary fix.
Internally, Apple apparently calls the Google-powered Siri upgrade by the code-name “Linwood,” according to the report, and rather than announcing an exciting AI partnership, Apple regards Google’s role in this deal as that of a “behind-the-scenes technology supplier,” Gurman writes.
Apple’s AI assistant, released in 2010, has long been a perceived weakness in Apple’s mobile software lineup. Even though it has been tweaked since the release of the Apple Intelligence suite, the voice-activated robot voice never became the centerpiece of some futuristic new AI-driven version of the iPhone as Apple once signaled it would. Instead it has come to symbolize Apple’s sluggishness in the AI race—mostly perceived as a 15-year-old piece of software that tells you the weather, and says it doesn’t understand you or can’t help you aside from that.
According to Gurman’s past reporting, the upgrade will be in place this coming spring as part of the update to iOS 26.4. This round of changes to Siri will be focused on “personalization,” allowing the assistant “to tap into consumers’ personal data and on-screen activities in order to better fulfill queries.”
According to MacRumors, the upgrade will still involve Apple’s proprietary models when queries can be answered with a lower-powered model that runs on the device itself, and there will be privacy measures in place for when Google’s software gets involved. “User privacy will be preserved by running the Google models on Apple’s server infrastructure without any external data sharing, and on-device personal data will likely be processed using Apple’s own Foundation Models,” writes Benjamin Mayo of MacRumors.
Gurman says the switch to Google Gemini will occur when Siri needs to perform complex tasks or summarize information.
Gurman’s report, as usual, comes from anonymous leaks, while Apple’s formal statement in response to this report notes that the spring release of new Siri features is not something the company has formally announced.
As Gizomodo’s James Pero wrote in July of this year, Apple was reportedly in talks with OpenAI and Anthropic about the possibility of powering this Siri upgrade with a GPT or Claude model. Given that the AI announcements in July didn’t match what had been previously advertised, it looked like things were “not going as planned.”
Now we have one more data point hinting at what’s been going on behind the scenes at Apple’s AI workshop. Apple supposedly still intends to develop its own AI system to power some future version of Siri, even as AI staff defections mount—for example, Ke Yang, who was spearheading Apple’s in-house chatbot effort, and who left for Meta last month. But in the meantime, it seems a $1 billion-per-year Google Gemini model will do.
