As we speak, we’re enhancing Android improvement with Gemma 4, our newest state-of-the-art open mannequin designed with complicated reasoning and autonomous tool-calling capabilities.
Our imaginative and prescient is to allow native agentic AI on Android throughout the complete software program lifecycle, from improvement to manufacturing. Android helps a spread of Gemma 4 fashions, from essentially the most environment friendly ones working straight on-device in your apps to extra highly effective ones working in your improvement machine that can assist you construct apps. We’re bringing Gemma 4 to Android builders by way of two pillars:
- Native-first Agentic coding: Expertise highly effective, native AI code help with Gemma 4 in Android Studio in your improvement laptop.
- On-device intelligence: Construct clever experiences utilizing the ML Package GenAI Immediate API to run Gemma 4 straight on Android system {hardware}.
Coding with Gemma 4 in Android Studio
When constructing Android apps, Android Studio can use Gemma 4 to leverage its state-of-the-art reasoning energy and native help for software use, whereas conserving the mannequin and inference contained fully in your native machine.
Gemma 4 was educated on Android improvement and designed with Agent Mode in thoughts. Because of this when you choose Gemma 4 as your native mannequin, you possibly can leverage the complete suite of Agent Mode capabilities for quite a lot of Android improvement use instances, together with refactoring legacy code, constructing a complete app or new options, and making use of fixes iteratively.
Be taught extra in regards to the potentialities Gemma 4 brings to your app improvement move and how one can get began.
Prototyping with Gemma 4 on-device
Because the introduction of Gemini Nano as the muse mannequin on Android, it has turn out to be accessible on over 140 million units. Gemma 4 is the bottom mannequin for the subsequent era of Gemini Nano (Gemini Nano 4) that’s optimized for efficiency and high quality on Android units. This mannequin is as much as 4x quicker than the earlier model and makes use of as much as 60% much less battery.
To make it as simple as potential to preview and prototype with Gemma 4 E2B and E4B fashions straight on AICore-supported units, we’re launching the AICore Developer Preview. Whereas we proceed to develop the ML Package GenAI Immediate API floor to unlock extra superior capabilities of the mannequin, you possibly can already begin exploring new use instances with Gemma 4 utilizing the Immediate API.
Put together your apps for the launch of the Gemini Nano 4 on the brand new flagship Android units later this 12 months by prototyping with Gemma 4 in the present day. Learn in regards to the upcoming options and deep dive into AICore Developer Preview and its Gemma 4 help right here.
Native agentic intelligence with Gemma 4
Working Gemma 4 domestically, you possibly can leverage its superior reasoning and tool-calling capabilities in your total workflow, from growing with the AI coding assistant in Android Studio to transport clever options in your app with ML Package GenAI Immediate API. This local-first method, accessible underneath Gemma’s open Apache license, gives another for builders to innovate in a privacy-centric and price efficient method. In a future launch, we are going to replace Android Bench to incorporate Gemma 4 and different open fashions, offering the quantified information you want to navigate efficiency trade-offs and choose the most effective mannequin on your use case.
We will’t wait to see what you construct!