Editorial illustration for Google, Arm, and Qualcomm boost Gemma 4 AI to run 4× faster on Android
Gemma 4: 4x Faster AI on Android Devices
Google, Arm, and Qualcomm boost Gemma 4 AI to run 4× faster on Android
Why does this matter? Because putting a capable language model on a phone has always meant a trade‑off between speed and battery life. Gemma 4 promises to break that pattern, offering a locally‑run AI that never sends user data off the device.
Yet the promise only holds if the software can keep up with the hardware people actually carry. Google has therefore leaned on two of the chip world’s biggest names to tune the model for today’s smartphones. The result, according to the companies, is a version of Gemma 4 that feels markedly quicker than its predecessor while sipping far less power.
Early numbers suggest the improvement isn’t just incremental; the gains could reshape how developers think about on‑device intelligence. Below, Google lays out the performance claims and Arm’s own benchmark figures that hint at an even larger leap.
Google also teamed up with Arm and Qualcomm to optimize the phone variants for current mobile chips. According to Google, Gemma 4 on Android runs up to four times faster than the previous generation while cutting battery drain by up to 60 percent. Arm's own benchmarks show even bigger gains: an aver
Google also teamed up with Arm and Qualcomm to optimize the phone variants for current mobile chips. According to Google, Gemma 4 on Android runs up to four times faster than the previous generation while cutting battery drain by up to 60 percent. Arm's own benchmarks show even bigger gains: an average 5.5x speedup in processing, provided the device packs a newer Arm chip with the SME2 instruction set, an extension that accelerates matrix math for AI models directly in silicon.
Agent skills bring tool use to on-device AI The app requires Android 12 or iOS 17. The two phone-sized variants differ in RAM requirements: E2B uses about 1.3 GB quantized and runs on devices with 6 GB of RAM, while E4B needs around 2.5 GB of model memory and at least 8 GB of RAM. Beyond basic chat, image recognition, and audio transcription, the app ships with what Google calls "agent skills": Wikipedia search, interactive maps, auto-generated summaries, and flashcards.
Gemma 4 can also describe photos, turn spoken input into diagrams and visualizations, and even team up with other local models for things like text-to-speech or image generation.
Will your phone feel faster? Google says Gemma 4 can run entirely on‑device, handling text, images and audio without sending data off the handset. The model also claims autonomous access to tools such as Wikipedia, interactive maps and QR‑code generators via built‑in agent skills.
On lower‑end devices, the E2B and E4B variants operate with just 6 GB or 8 GB of RAM, yet Google reports up to four‑times the speed of the previous generation. Battery usage, according to the company, drops by as much as 60 percent. Arm and Qualcomm have worked with Google to tune the software for current mobile silicon, and Arm’s own benchmarks suggest even larger performance gains.
However, the figures come from the manufacturers; independent testing on a broader range of Android hardware has not been published. It remains unclear whether real‑world usage will match the advertised improvements, especially under varied network conditions or with third‑party apps. The promise of fully private, on‑device AI is notable, but its practical impact will depend on how consistently those performance and efficiency claims hold up across the fragmented Android ecosystem.
Further Reading
- Gemma 4 on Arm: Accessible, immediate, optimized on-device AI to accelerate the mobile app experience - Arm Newsroom
- Gemma 4: The new standard for local agentic intelligence on Android - Android Developers Blog
- Gemma 4: Our most capable open models to date - Google Blog
- Announcing Gemma 4 in the AICore Developer Preview - Android Developers Blog
Related Reading
Browse all Google news →
Open Source UK PM vows action on Grok's deepfake scandal, Starmer condemns X
Open Source GPT-5 helps mathematicians offload tedious tasks, says Timothy Gowers
Open Source India proposes licensing and royalty rules for AI training by Google, OpenAI
Related Topic Gemini 3 Pro builds screenshot-to-code app in two prompts, fixes bugs
LLMs & Generative AI
Related Topic Gemini 3 Pro and GPT-5 stumble on graduate-level physics benchmark
LLMs & Generative AI
From Archives Operator calls AI agent that defamed Shambaugh a 'social experiment
From Archives NVIDIA launches AITune v0.2.0 with KV‑cache support for LLM inference
From Archives Google AI's PaperOrchestra boosts manuscript success, 79‑81% win rate
Research & Benchmarks
From Archives Google Gemini AI generates real-time 3D simulations, like Moon orbiting Earth
LLMs & Generative AI
Common Questions Answered
How much faster does Gemma 4 run on Android devices compared to its previous generation?
According to Google, Gemma 4 runs up to four times faster on Android devices than its previous version. Arm's benchmarks even suggest an average 5.5x speedup in processing, particularly on newer Arm chips with the SME2 instruction set.
What battery performance improvements does Gemma 4 offer on mobile devices?
Google reports that Gemma 4 cuts battery drain by up to 60 percent compared to previous generations. This significant battery efficiency improvement allows the AI model to run more sustainably on mobile devices without compromising performance.
What unique capabilities does Gemma 4 offer for on-device AI processing?
Gemma 4 can run entirely on-device, handling text, images, and audio without sending data off the handset. The model also provides autonomous access to tools like Wikipedia, interactive maps, and QR-code generators through built-in agent skills, all while operating efficiently on devices with just 6-8 GB of RAM.