Apple and Google’s AI partnership may be much deeper than outsiders know, according toThe InformationAccording to the latest report, Apple has gained “far more freedom than expected” in this multi-year cooperation agreement. Not only can it fully access the Gemini model, but it can also modify and “distillate” it in its own data center to create a lightweight model more suitable for Siri.
Apple gains full access to Gemini
According to “The Information”‘s “AI Agenda” column, Apple and Google’s AI collaboration is “deeper than previously known.” The agreement shows that Apple has full access to the Gemini model and can use these permissions within its own data center facilities.
Specifically, Apple can:
- Full access to Gemini models within your own data center
- Use these access rights to produce smaller specialized models
- Build lightweight models that run directly on Apple devices
- Make specific tasks faster and more efficient
“Knowledge distillation” technology: making large models slim down
The report pointed out that the technology Apple is using is called “knowledge distillation” (distillation). This is a way for a large language model (LLM) to transfer knowledge to a smaller, leaner model.
The advantages of this approach are:
- Small models are more efficient
- Requires less computing power
- Respond faster
- More suitable for execution on the device rather than relying on the cloud
Because Apple has full access to Gemini, its “student model” can not only learn to imitate Gemini’s output answers, but also learn the calculations used in Gemini’s internal operations. This approach is more effective than simply imitating the output results, and can produce a small model with performance close to that of the advanced teacher model, but with significantly reduced computing requirements.
Multi-year partnership worth $1 billion
This collaboration was first officially announced in January 2026, when Apple and Google signed a multi-year cooperation agreement to use Google’s Gemini models and cloud technology to build future Apple Foundation Models.
Apple said in a statement: “After careful evaluation, we determined that Google’s technology provides the strongest foundation for Apple Foundation Models, and we are excited about the innovative experiences it will unlock for users.” According to previous reports, the deal is worth approximately $1 billion per year.
Apple 確認與 Google 合作，將推出基於 Gemini 驅動 Apple Intelligence 與 Siri
It is worth noting that this cooperation is not an exclusive agreement. Apple also tested technology from competitors such as OpenAI and Anthropic before settling on Google.
Major Siri update expected to debut at WWDC
Apple plans to unveil major revisions to Siri at its WWDC developer conference in June this year, including:
- Siri can remember past conversations with users
- Proactive suggestions (e.g. reminding users to leave early to avoid airport pick-up traffic)
This is probably the Siri vision (big pie) planned at the time of WWDC 2024. Although a few years have passed and two generations of iPhone have not yet been completed (and have been sued for this), this time Apple may really have a chance to succeed with the help of Google Gemini.
Challenges and limitations
Despite the high degree of freedom, sources noted that the process still has challenges. Apple’s goals for Siri don’t always align with Gemini’s areas of expertise, which can lead to friction in the model distillation process. In addition, the report also reiterated that Apple’s Foundation Models (AFM) team has not given up on internal model development. However, the specific goals of the AFM team still seem unclear, showing that Apple is still looking for a balance between independent research and development and external cooperation.
This collaboration between Apple and Google reveals an important trend: Even technology giants with strong R&D capabilities are choosing cooperation over total autonomy in the AI race. For Apple, this deal not only makes up for Siri’s long-term lagging shortcomings, but also gives it rare freedom to modify models; for Google, it is a major victory for the Gemini ecosystem and may shake OpenAI’s leadership in the consumer AI market. In terms of results, although Apple did not come up with its own model in the end and Siri also used Gemini as the core, it did not need to spend tens of billions of dollars to train iterative models like other AI companies. It only needed to spend one billion dollars to get everything done. It is no wonder that some commentators say that Apple is the winner with the biggest profit (least expenditure) under this wave of AI.
Source: KOCPC Chinese