Copyright Wccftech
Apple is increasingly relying on Google's crutches to power its revamped Siri in the cloud. And we now know some of the specs of the custom Gemini model that is expected to bolster the new Siri. Mark Gurman: Apple is planning to use a custom Gemini AI model with 1.2 trillion parameters to power the revamped Siri Bloomberg's Mark Gurman is now reporting that Apple is planning to use a gigantic, albeit tailored, Gemini AI model to power its upcoming revamped Siri. With 1.2 trillion parameters under its belt, the customized Gemini model would "dwarf" the 1.5 billion-parameter, bespoke AI model that Apple currently uses to power Siri in the cloud. Apple tested OpenAI's ChatGPT and Anthropic's Claude LLMs before settling on a customized version of Google's Gemini to process the more complicated user requests that are to be offloaded to the cloud under Apple's Private Cloud Compute framework, which will safeguard user privacy by using encrypted and stateless data. According to Gurman, Apple will pay Google around $1 billion per year to use Google's proprietary AI technology, as per the terms of the agreement that is currently in the process of being finalized. This forms merely the latest tranche in the ongoing transactional relationship between the two tech giants. After all, Google already pays Apple $20 billion per year for ensuring default search engine privileges within the Safari browser and across other Apple services. Coming back, Apple has even given an internal codename - Glenwood - to its ongoing efforts to fix Siri, largely with the help of third-party models. These efforts are currently being spearheaded by Vision Pro headset creator, Mike Rockwell, and software engineering chief, Craig Federighi. Do note that the upcoming Siri revamp will have three major components: Query planner - this is Siri's decision-making layer that decides on the most efficient route to fulfill a given user request, with available options including web search, accessing personal data such as calendar entries or photos, or using a third-party app via App Intents, which itself is a framework that makes a given app discoverable and usable by Siri, allowing users to perform certain in-app actions without having to physically open the app. Knowledge search system - Siri will gain a general knowledge database to answer trivia queries without having to resort to ChatGPT, other third-party AI integrations, or web-based results. Summarizer - this is a core tool within Apple Intelligence, allowing Siri to leverage third-party AI models such as ChatGPT to summarize a given text or audio snippet, and includes notification summaries, Safari webpage summaries, writing tools, etc. Under Apple's evolving architecture, the customized Gemini model from Google will handle Siri's query planner and summarizer functions, leaving the Knowledge search system to Apple's on-device LLMs. Critically, this partnership will not place Google's search AI at an apex position within Apple's sprawling ecosystem. Of course, Apple does not intend to use Google's Gemini model indefinitely, and is still working on developing an in-house solution. In the interim though, the Gemini model will function as a critical crutch for Apple to launch the revamped Siri with iOS 26.4, bringing in-app actions, personal context awareness, and on-screen awareness to its bespoke voice assistant.
