Google has announced details for Gemini Nano 4 and Gemma 4, with Gemma 4 serving as the foundational technology for the next generation of Gemini Nano models for Android.
Gemini Nano 4 is currently accessible to developers in early access through the AICore Developer Preview.
Model Overview and Availability
Google previously referred to these models as "nano-v2" and "nano-v3." The "nano-v3" iteration is currently available on devices such as the Pixel 10 and Galaxy S26. Gemini Nano 4 is scheduled to launch on new flagship Android devices later in the current year, promising significant advancements in on-device AI.
Two primary versions of Gemini Nano 4 have been detailed for developers:
- Gemini Nano 4 Fast [Preview, TPU]:
This version is based on Gemma 4 E2B. It is designed for rapid responses, offering speeds reported to be up to three times faster than the E4B model and lower latency. - Gemini Nano 4 Full [Preview, TPU]:
This version is based on Gemma 4 E4B. It is designed for enhanced reasoning capabilities, suitability for complex tasks, and aims to provide higher-quality results.
Performance and Capabilities for Android
For Android devices, Google reports that Gemma 4 offers several significant performance enhancements. These include:
- Operates up to four times faster than preceding versions.
- Consumes up to 60% less battery power.
- Provides multimodal understanding across text, image, and audio input.
- Natively supports over 140 languages.
Key Use Cases
The new models are designed to enhance various applications and tasks across Android devices:
- Reasoning: Offers improved quality for chain-of-thought commands and conditional statements. This includes critical tasks such as determining if a comment adheres to community guidelines based on criteria like profanity, derogatory language, or hate speech.
- Math: Provides enhanced accuracy for mathematical problem-solving, including complex calculations like periodic contributions to a savings goal.
- Time Understanding: Demonstrates greater capability in reasoning about time, significantly improving features related to calendars, reminders, and alarms. An example cited is scheduling a reminder a set duration before an event.
- Image Understanding: Delivers more accurate results for Optical Character Recognition (OCR) tasks, encompassing chart analysis, visual data extraction, and handwriting recognition.
Developer Access and Future Updates
Developers can register for the AICore Developer Preview to gain early access to Gemini Nano 4. A significant advantage for developers is that code developed for Gemma 4 is designed to be automatically compatible with Gemini Nano 4-enabled devices.
Google plans to introduce further updates during the preview period. These enhancements will include crucial features such as support for tool calling, structured output, system prompts, and a thinking mode within the Prompt API.