Google is back on its feet: it will implement Gemini Nano also on the Pixel 8

Earlier this month, during a session of #TheAndroidShow, Google declared that its artificial intelligence would not arrive on the Pixel 8 due to “hardware constraints”.
This despite the fact that, going back a few more months, at the launch of Gemini AI, there was discussion of a model also designed for run on mid-range smartphones.

Now, once again turning the tables, Google has announced that model Gemini Dwarf it will also soon be available on Pixel 8 through an upcoming update.

This turnaround was made known via a post on Instagram from the official Made by Google profile.

RAM differences

Gemini Nano is a smartphone-optimized version of the Gemini multi-modal artificial intelligence model.
It is currently available on devices like the series Galaxy S24 by Samsung e Pixel 8 Proincluding features like summary in the Google Recorder app and Magic Compose smart responses.

The decision to initially limit the Gemini Nano to the Pro model, despite the Pixel 8 sharing the same Tensor G3 processor, was mainly attributed to the difference in the amount of RAM available: 8GB for the Pixel 8 and 12GB for the Pixel 8 Pro.
Google explained that running complex language models on devices with different memory specifications can impact user experience.

At the end of the fair

The Gemini Nano AI model will soon be introduced as a developer preview via an upcoming Pixel Feature Drop.
It could happen as early as June 2024 Feature Dropaccompanied by Android 14 QPR3, the new features of which are already available in preview for members of the beta program.

Google seems to have found a solution to mitigate the impact of Gemini Nano on RAM consumption and overall performance.
Finally, it must be said that even Samsung's Galaxy S24, which has 8 GB of RAM, has already benefited from the capabilities of Gemini Nano since its launch.
This indicates that Google may have actually been driven by the choice to ensure an optimal user experience by avoiding overloading the device hardware with an artificial intelligence model, especially when there is a cloud-based alternative for AI functions.