Overview
- Bloomberg reports Apple is finalizing an agreement to pay roughly $1 billion annually for access to a custom Google Gemini model to power key parts of the next Siri.
- The Google model is described at about 1.2 trillion parameters, far larger than Apple’s current cloud model, and uses a Mixture‑of‑Experts design that activates only a subset of parameters per request.
- Under the reported plan, Gemini would handle Siri’s summarization and multi‑step planning, while other Siri features would continue to rely on Apple’s in‑house models.
- The Gemini system would run within Apple’s Private Cloud Compute, with reporting indicating Google would not receive user query data.
- Apple is still developing its own large cloud model targeting around 1 trillion parameters and is aiming to roll out the overhauled Siri in spring 2026, after evaluating options from OpenAI and Anthropic and favoring Google in part on price.