Gemini Intelligence with high hardware requirements for smartphones

Even during the announcement of Gemini Intelligence during the Android Show I/O Edition, Google hinted with the sentence “The best of Gemini on our most advanced devices” that on-device AI will have high demands on devices. Google has left concrete details in the footnotes: According to them, the Pixel 9 series and even last year's Galaxy Z Fold 7 are out of the running – apparently even the Xiaomi 17 Ultra.

“Highest performance and system requirements”

“Gemini Intelligence features are only available on Android devices with the most advanced capabilities and spec requirements,” writes Google in the footnote and specifies that devices must have at least 12 GB of RAM. This means that most entry-level and mid-range devices are already out; even Google's own A-series, which has 8 GB of RAM, cannot run Gemini Intelligence.

Another hurdle is the processor: Google only writes “qualified SOC” here and in parentheses “flagship chip.” The manufacturer likely means Qualcomm's and MediaTek's top models, as well as the company's own Tensor chips of the current and upcoming generation.

But that's not all: Google also sets the bar high for the required on-device AI version: at least model Gemini Nano v3 or newer must be supported. This is where the wheat is separated from the chaff, even among the more powerful models: on the developer website for GenAI APIs from ML Kit, Google lists which devices are compatible with v3 and which remain on the older version 2.

Only smartphones from 2025 and newer are compatible with Gemini Nano v3 – this means, among other things, that Google's Pixel 9 series, Honor's Magic V5, Magic 7 and Magic 7 Pro, and the OnePlus 13 and others are out. But even devices from last year and even top-current devices are incompatible with Gemini Intelligence according to the list: for example, Samsung's Galaxy Z Fold 7, as well as the 2026 devices Xiaomi 17 and 17 Ultra, and a whole range of other Xiaomi models are incapable of running Gemini Intelligence.

Update guarantee, secure virtualization

Beyond the hardware and software requirements, Google makes clear specifications for Gemini Intelligence regarding updates: If manufacturers want to use the new AI on their devices, they must guarantee at least 5 years of Android updates and six years of quarterly security patches.

Furthermore, Google mandates the use of virtualization technologies announced last year, pKVM (protected kernel-based virtual machine), to secure the data processed by the AI. These are intended to guarantee that the information is processed only locally and that no unauthorized apps have direct access.

The advantage of Gemini Intelligence, by the way, is that the tasks are executed directly on the device, not in the cloud, which is a plus point in terms of privacy. Those who don't want to use AI don't have to: like the previous Gemini assistant, Gemini intelligence is likely to remain optional.

(afl)