Apple Taps Google Gemini for New Siri

The decision to integrate Gemini came after Apple evaluated models from multiple AI leaders, including OpenAI and Anthropic. Anthropic was reportedly a frontrunner until negotiations stalled over financial terms, with the company allegedly seeking several billion dollars annually. The multi-year deal with Google, valued at a reported $1 billion per year, leverages a long-standing partnership and provides Apple with a proven, scalable foundation model. This partnership addresses the significant delays and performance challenges that hampered Apple's internal efforts to modernize Siri. By opting for Google's established models, Apple is pragmatically accelerating its timeline to achieve parity with competing AI assistants, a move CEO Tim Cook has described as a "collaboration" to secure the "most capable foundation" for Apple's own models. The technical execution hinges on a hybrid approach to maintain privacy. Simpler AI tasks will continue to run on-device via Apple's Foundation Models, while more complex requests are routed to Private Cloud Compute (PCC) servers, not directly to Google's cloud. This architecture allows Apple to utilize powerful, server-based models for intensive tasks without third-party data retention. Apple's Senior VP, Craig Federighi, detailed the novel security of the PCC infrastructure, describing it as a "hermetically sealed privacy bubble." The custom Apple Silicon servers are designed without persistent storage, meaning no user data is retained or logged after a request is processed, making it cryptographically unrecoverable and inaccessible even to Apple employees. The mention of a 1-million token context window refers to a key capability of Google's Gemini 1.5 Pro. This massive context window allows the model to process and understand vast amounts of information in a single prompt—equivalent to over 700,000 words or 30,000 lines of code—enabling the promised on-screen awareness and complex, multi-app task chaining.