- Google has announced “Gemini Nano 4 (Gemma 4),” the next-generation model of its on-device AI “Gemini Nano,” and confirmed it will debut on the upcoming next-generation Google Pixel devices arriving later in 2026.
- The next-generation “Pixel 11” series will come with “Gemini Nano 4” pre-installed.
- The key highlight of “Gemini Nano 4” is its dramatic increase in efficiency: compared to the previous “Gemini Nano,” inference speed has improved by up to 4x, with a reduction in battery consumption of up to 60%.
On Thursday, April 2, 2026, Google announced “Gemini Nano 4 (Gemma 4),” the next generation of its on-device AI model “Gemini Nano,” while also revealing that it will be featured on the next-generation Google Pixel devices launching later this year.
Currently, the “Gemini Nano” on-device AI model powers over 140 million Android devices. However, the next-generation Google Pixel devices, arriving in late 2026, will be built on “Gemini Nano 4,” which is based on “Gemma 4,” the latest multimodal iteration of the lightweight yet high-performance “Gemma” series of small language models. This means that the next-generation “Pixel 11” series will ship with “Gemini Nano 4” pre-installed.
The standout feature of the “Gemma 4”-based “Gemini Nano 4” is its dramatic improvement in efficiency. Compared to its predecessor, it delivers up to 4x faster inference speeds and cuts battery consumption by up to 60%.
Furthermore, “Gemini Nano 4” will be available in two distinct models tailored to specific use cases: “E4B,” a full-spec model specialized for advanced reasoning and complex task processing, and “E2B,” a model highly optimized for speed and low latency, running up to 3x faster than the E4B. It also features native support for over 140 languages.
Additionally, “Gemini Nano 4” represents a significant evolution in functionality through its full-scale support for multimodal processing, allowing it to understand and handle not just text, but also images and audio. Optical Character Recognition (OCR) precision has seen a massive boost, enabling more accurate on-device chart reading, handwriting recognition, and visual data extraction.
On top of that, the model’s fundamental reasoning capabilities have been greatly enhanced. It now supports chain-of-thought reasoning and processing with complex conditional logic, while its mathematical proficiency has improved to handle everything from daily calculations to calculating savings goals with high accuracy. It also possesses a deeper understanding of the concept of time, promising more precise assistance for tools like calendars and alarms, such as setting reminders by counting backward from an event date.
While “Gemini Nano 4” will be officially released alongside the “Pixel 11” series later this year, a preview version is already available through the “AICore Developer Preview” program. Interested users can access it by joining the “aicore-experimental” Google Group and registering as a “Beta Tester” for the “Android AICore” app on the Google Play Store to receive the beta build.
Source：Google
コメントを残す