Gemma 4 Download to Mobile
Put Gemma 4 on your iPhone or Android phone and use it with no connection. One free app from Google, one model download, and everything runs on the device.
iOS 17 or later · Android 12 or later · free app from Google ·
Same app, two stores
Google AI Edge Gallery is Google's own free app for running models on your device. Gemma 4 is downloaded from inside it.
iPhone and iPad
- Requires
- iOS 17.0 or later
- App size
- 140.9 MB
- Price
- Free
- Also runs on
- Mac with an M1 chip or later (macOS 14), Apple Vision
- Seller
- Google LLC
Android phones and tablets
- Requires
- Android 12 or later
- Price
- Free
- Direct file
- APK on GitHub releases
- Source
- Open source on GitHub
- Publisher
Details from the App Store listing and Google's AI Edge Gallery repository, October 7, 2026. The app offers AI Chat with a thinking mode, Ask Image, Audio Scribe, Prompt Lab, Agent Skills, Mobile Actions and model management. Android has its own in-depth Android page.
Four steps, about ten minutes
The steps are the same on iPhone and Android. Most of the time is the model download.
Install Google AI Edge Gallery
From the App Store or Google Play. Check the publisher is Google before installing.
Download a model on Wi-Fi
Open the app, choose Gemma 4 E2B (2.58 GB) or E4B (3.66 GB) and download it inside the app. Keep the app open until it finishes.
Choose the GPU backend
In the model's settings, select GPU where it is offered. It is the biggest single speed-up - see the numbers.
Test it in airplane mode
Turn the network off and ask a question. If it answers, everything is running on your phone.
On a computer instead? Start at the download hub, how to download or the install guide.
E2B or E4B?
Only the two edge models are built for phones. Start with E2B; move to E4B on a recent flagship if you want better answers more than faster ones.
| Gemma 4 E2B | Gemma 4 E4B | |
|---|---|---|
| Download | 2.58 GB | 3.66 GB |
| Writing speed, iPhone 17 Pro | 56.5 tokens a second | 25.1 tokens a second |
| Writing speed, Galaxy S26 Ultra | 52.1 tokens a second | 22.1 tokens a second |
| Wait for the first word | 0.3 s | 0.8 - 0.9 s |
| Inputs | Text, image, audio | Text, image, audio |
| Context in the phone build | Up to 32K tokens | Up to 32K tokens |
| Knowledge score, MMLU Pro | 60.0 | 69.4 |
| Best for | Most phones; quick replies | Recent flagships; harder questions |
Speeds are Google's GPU measurements from the E2B and E4B LiteRT-LM model cards; your phone will differ. The models themselves accept up to 128K tokens, but the phone builds list a 32K maximum. Quality differences are in the benchmarks and model comparison; full details in technical specs and on the models page.
iPhone vs Android
On current flagships the two are close. Writing speed is slightly higher on the iPhone; reading speed is higher on the Galaxy; the wait for the first word is the same.
Gemma 4 E2B - writing speed
Gemma 4 E4B - writing speed
| Phone and backend | Model | Reading | Writing | First word | Memory |
|---|---|---|---|---|---|
| iPhone 17 Pro · GPU | E2B | 2,878 t/s | 56.5 t/s | 0.3 s | 1,450 MB |
| Galaxy S26 Ultra · GPU | E2B | 3,808 t/s | 52.1 t/s | 0.3 s | 676 MB |
| iPhone 17 Pro · CPU | E2B | 532 t/s | 25.0 t/s | 1.9 s | 607 MB |
| Galaxy S26 Ultra · CPU | E2B | 557 t/s | 46.9 t/s | 1.8 s | 1,733 MB |
| iPhone 17 Pro · GPU | E4B | 1,189 t/s | 25.1 t/s | 0.9 s | 3,380 MB |
| Galaxy S26 Ultra · GPU | E4B | 1,293 t/s | 22.1 t/s | 0.8 s | 710 MB |
| iPhone 17 Pro · CPU | E4B | 159 t/s | 9.7 t/s | 6.5 s | 961 MB |
| Galaxy S26 Ultra · CPU | E4B | 195 t/s | 17.7 t/s | 5.3 s | 3,283 MB |
Use the GPU on both
It cuts the wait for the first word from about 2 seconds to 0.3 on E2B, and from 5 - 6 seconds to under 1 on E4B.
Memory behaves differently
In Google's tests the GPU backend used less memory than CPU on the Galaxy, but more on the iPhone - 3,380 MB for E4B. That makes E4B a tighter fit on iPhones.
A drafter can add speed
With speculative decoding on the Galaxy's GPU, Google measured E2B at 66 - 92 tokens a second, up from 51.5, and E4B at 37 - 49, up from 21.9, depending on the task.
"Reading" is prefill and "writing" is decode, in tokens per second, with 1,024 tokens in and 256 out. From Google's E2B and E4B LiteRT-LM model cards. These are flagship phones; older and mid-range devices will be slower. How the drafter works is on the capabilities page.
What your phone needs
A recent system
iOS 17 or later, or Android 12 or later - the versions Google lists for the app.
Free storage
2.58 GB or 3.66 GB for the model, plus the app. Aim for about 6 GB free for E2B and 8 GB for E4B.
Enough memory
In Google's tests E2B used 0.6 - 1.7 GB while running and E4B 0.7 - 3.4 GB, depending on phone and backend.
Battery and heat
Long answers drain the battery and warm the phone faster than normal use. When idle it costs nothing.
The storage suggestion is this site's guidance, not a Google figure. Google does not publish a list of supported phone models; if the app installs and the model loads, it runs. For computers see the VRAM guide.
What stays on your phone
Your prompts, photos and audio are processed on the device. The app itself still reports some usage data, as its store listings disclose.
Your conversations
Google's repository says processing happens "directly on your device hardware" and that no internet is required.
App Store label
Lists a device identifier and diagnostics as linked to you, and coarse location, product interaction and crash data as not linked to you.
Google Play label
Says the app collects app activity and performance data and shares none with third parties.
Store labels as read in October 2026; they can change with app updates. For the strictest privacy, use the app in airplane mode. Responsible-use notes are under safety.
Other apps that run Gemma on a phone
AI Edge Gallery uses Google's .litertlm format with GPU acceleration. Community apps mostly use GGUF files: more models to choose from, often slower.
| App | Platform | Format | Best for |
|---|---|---|---|
| Google AI Edge Gallery | iPhone and Android | .litertlm | Easiest setup and the best acceleration |
| PocketPal AI | iPhone and Android | GGUF | Open source; switching between models |
| Private LLM | iPhone | Its own builds | A polished, paid app |
| LLM Farm | iPhone | GGUF | Open source and configurable |
| Termux with Ollama | Android | GGUF | Command-line users |
The community apps are carried over from this page's earlier version and were not re-tested for this update - check that each lists Gemma 4 before relying on it. GGUF files are covered under QAT builds. You can also skip installing anything and use a hosted option from your phone's browser: see chat, the playground and the API page.
Download the phone builds directly
The same files the app uses are on Hugging Face, for building Gemma 4 into your own iOS or Android app with LiteRT-LM.
| Build | Repository | Size | Use |
|---|---|---|---|
| Gemma 4 E2B | litert-community/gemma-4-E2B-it-litert-lm | 2.58 GB | Chat, vision and audio on most phones |
| Gemma 4 E4B | litert-community/gemma-4-E4B-it-litert-lm | 3.66 GB | Higher quality on recent flagships |
| EmbeddingGemma 2, text | litert-community/embeddinggemma-2-text-270m-litert-lm | 165 MB | On-device search over text |
| EmbeddingGemma 2, full | litert-community/embeddinggemma-2-740m-litert-lm | 485 MB | Search across text, images, audio and video |
# Try the same phone build on a computer first
pip install -U litert-lm
litert-lm run --from-huggingface-repo=litert-community/gemma-4-E2B-it-litert-lm --prompt="Write me a poem"
Command from the LiteRT-LM model card. Runtime and source: LiteRT-LM overview · google-ai-edge/LiteRT-LM. Android-specific routes, including AICore, are on the Android page; embeddings are in the EmbeddingGemma 2 guide; usage tips in prompting and features.
The bigger models need a computer
The 12B, 26B A4B and 31B do not fit on phones. If you want their quality from a mobile device, run one on a computer or use a hosted service.
Gemma 4 26B A4B and 31B
14.4 GB and 17.5 GB at 4-bit. For a 24 GB GPU or a 32 GB Mac. Choose with the runtime selector.
Reach it from your phone
Host a model yourself with the self-hosting guide, or pay per token - see pricing.
Is there an official Gemma 4 app for phones?
The official route is Google AI Edge Gallery, a free app from Google on the App Store and Google Play. Gemma 4 is downloaded inside it. There is no separate "Gemma" chatbot app.
Can I download Gemma 4 on an iPhone?
Yes. Install Google AI Edge Gallery, which needs iOS 17 or later, then download Gemma 4 E2B or E4B in the app.
Does it work without internet?
Yes, completely, once the model has downloaded. Turn on airplane mode to check.
Is it free?
Yes. The app is free and the models are open under Apache 2.0. There is no subscription and no per-message charge.
Which is faster, iPhone or Android?
On current flagships they are close: 56.5 against 52.1 tokens a second on E2B, with the same 0.3 second wait for the first word. It should not decide which phone you use.
How much storage do I need?
2.58 GB for E2B or 3.66 GB for E4B, plus the app. Leaving about 6 to 8 GB free gives comfortable room.
Can Siri or Apple Intelligence use Gemma 4?
No. Gemma 4 runs inside the app you install it in; it does not replace the phone's built-in assistant.
How good is a phone-sized model?
Good for summaries, questions, drafting, photos and voice notes; weaker than large models on hard reasoning and long tasks. See limitations and, for the previous on-device generation, the Gemma 3 page.
Put Gemma 4 in your pocket
Install the app, download a model on Wi-Fi and try it in airplane mode.