On May 26 Beijing Time, according to reports from multiple media outlets citing informed sources, Apple is not simply integrating Gemini into Siri, but is instead using a customized 1.2 trillion parameter large language model developed by Google as the "brain" for the next generation of Siri's major overhaul.

This scale far exceeds current mainstream mobile models and has attracted significant attention in the industry.

Model Scale: 1.2T vs. Gemini 3.5 Flash 300B

It is estimated that Gemini 3.5 Flash has around 300 billion parameters, while the customized model Apple is using reaches 1.2 trillion parameters, which is significantly larger. AIbase analysis points out that if this massive model can be efficiently deployed, it will bring stronger understanding, reasoning, and complex task handling capabilities to Siri, especially achieving a qualitative leap in multimodal interaction and context understanding.

Performance and Speed: Local Response Is the Biggest Challenge

Although the number of model parameters has increased dramatically, Apple has always emphasized user privacy and real-time performance. The report emphasizes that simple queries are expected to run on local devices first. This means Apple must solve the challenge of efficient inference of large models on terminals like iPhones—ensuring answers to daily questions are fast enough, while also balancing power consumption and heat control.

AIbase believes that a model being "large" does not necessarily mean it is "good." In mobile scenarios, the balance between latency, energy consumption, and accuracy is the key to success. Whether Apple can achieve efficient local or hybrid deployment on the 1.2T parameter model will directly determine the user experience of this Siri overhaul.

The AI Battle Heats Up in the Second Half of the Year

As Apple prepares to showcase the deep integration of Apple Intelligence and Gemini at WWDC, the global competition among AI giants has entered a new phase. The following major updates are worth looking forward to in the coming months:

  • WWDC: Apple Intelligence makes its full debut, with Siri combined with the customized Gemini model
  • GPT-5.6: Progress on OpenAI's next-generation model
  • Sonnet 4.8 / Opus 4.8: Anthropic may release an update simultaneously
  • Gemini 3.5 Pro: Google has confirmed it will soon be released

AIbase will continue to track Apple's Siri upgrade progress and the practical application of large models on the terminal side. This AI competition, defined by parameter scale, inference speed, and privacy protection, is getting closer to consumers' daily usage scenarios. Who will ultimately win, let's wait and see.