Google enables Gemini Nano 4 developer preview on Pixel 11

Accessing the new AI preview

Google has introduced a new toggle for the Pixel 11 series that allows users to experiment with upcoming artificial intelligence. The setting, titled "Enable AICore Developer Preview," is located deep within the system's developer options. It is specifically designed to unlock preview models and features for the Android AICore system.

Android AICore is the underlying framework that allows the Gemini Nano model to run directly on a device's hardware rather than in the cloud. By providing this preview, Google is giving users a way to see how the next generation of these models behaves in a real-world environment. The option is currently turned off by default to ensure standard users are not affected by experimental code.

Previously, getting access to these experimental features was a much more difficult process for the average person. Users had to join a specific Google Group called "aicore-experimental" and register as a beta tester. As jetstream.blog reported, the new update removes this barrier by placing the toggle directly in the Settings menu under System and Developer options.

Meet Gemini Nano 4

The primary focus of this update is Gemini Nano 4, which is the successor to the current on-device AI model. This new version is built upon the multimodal Gemma 4 model, which is capable of processing different types of information like text and images. It represents a significant jump in how the phone handles complex tasks locally.

One of the most notable improvements for the new model is its efficiency regarding speed and power. It is designed to offer up to 4x faster inference speeds, which refers to how quickly the AI can generate a response. Additionally, it aims for up to a 60% reduction in battery consumption, making it much more viable for long-term mobile use.

Because these models are designed to run on-device, they do not require an active internet connection to function. This allows for faster interactions and better privacy, as the data does not have to leave the phone to be processed. Gemini Nano 4 is intended to power these types of interactions while staying within the hardware limits of the Pixel 11 series.

Testing tools for developers

The update includes a specific test app that allows users to interact with the preview models in a controlled way. In this app, users can select different models and try out various prompts to see how the AI responds. It also provides several sliders and options to change the personality of the AI, such as its tone and style instructions.

Beyond just text, the test app supports multimodal inputs, meaning users can attach images or audio files. These can be pulled from the device's camera, gallery, or audio sources. This gives a clear picture of how the Gemma 4-based model handles different types of media in a single interaction.

For those who want to see what is happening under the hood, the app includes a debug info screen. This screen displays technical metrics such as the output rate per second and the time to first token. These numbers help developers understand exactly how the hardware is performing when running the new AI code.

Technical controls and risks

The developer preview offers several advanced configuration settings that are usually hidden from standard users. These include "Temperature," which controls the creativity of the AI's responses, and "Top-K," which limits the vocabulary the model chooses from. There is also a "Thinking" mode that allows the model to perform step-by-step reasoning before providing a final answer.

Users can also set a "System prompt" to give the AI specific instructions on how to behave during a conversation. These tools are essential for developers who are trying to fine-tune how an AI interacts with a specific app or service. It allows for a high level of customization that isn't available in the standard consumer version of Gemini Nano.

However, Google warns that because these are preview versions, they are not intended for daily use. The models might run slowly, provide inaccurate information, or cause the device to become unstable. They may also impact the phone's storage and overall performance, so users should be cautious when enabling these features.

The path forward for Pixel

This rollout marks a transition toward making high-performance AI more accessible to the developer community. By moving the preview into the standard developer options, Google is making it easier for people to experiment with the Gemma 4 architecture. This could lead to a wider variety of apps that use local AI for features like live translation or image editing.

The focus on battery efficiency and inference speed suggests that Google is preparing for a future where on-device AI is the standard. As the models become more capable while using less power, they will become less of a luxury and more of a core feature of the Android experience. The Pixel 11 serves as the primary testing ground for these capabilities.

Moving forward, the community will likely see more updates to the AICore framework as these models move out of preview. For now, the focus remains on gathering data from developers to ensure the models are stable enough for the general public. This staged rollout ensures that the most powerful tools are refined before they reach every user's pocket.