Google Pixel 9 Pro
Tensor G4, 16 GB RAM
Inference speed 0.8 s / run (NPU)
System requirements
Will Cove run on your phone?
Cove runs the AI model entirely on your device. To get smooth performance, your phone needs Android 12+, 4 GB RAM, and ideally a recent NPU. Here are the details.
Minimum vs recommended
Cove will install on the minimum spec, but inference will be slower (5–15 seconds per AI run instead of 1–3 seconds). For an iPhone-grade experience, aim for the recommended column.
| Spec | Minimum | Recommended |
|---|---|---|
| Android version | Android 12 (API 31) | Android 14+ (API 34) |
| RAM | 4 GB | 8 GB or more |
| Free storage | 5 GB free (model + cache) | 10 GB free (multiple apps + history) |
| SoC / NPU | Snapdragon 7 Gen 1 / Tensor G1 / Dimensity 8200 (no NPU OK) | Snapdragon 8 Gen 1+, Tensor G2+, Dimensity 9000+ (with NPU) |
| Screen | 5.0″ or larger | 6.1″+ for camera translation comfort |
| Camera | Any back camera (only required for Travel & Photo apps) | Autofocus, 48 MP+ for menu/sign translation |
Tested phones
We test every release on this set. If your phone is on the list, Cove will work — and we have data on real-world performance.
Google Pixel 8a
Tensor G3, 8 GB RAM
Inference speed 1.2 s / run (NPU)
Google Pixel 7
Tensor G2, 8 GB RAM
Inference speed 1.6 s / run (NPU)
Samsung Galaxy S24
Snapdragon 8 Gen 3, 8 GB RAM
Inference speed 1.0 s / run (NPU)
Samsung Galaxy S22
Snapdragon 8 Gen 1, 8 GB RAM
Inference speed 2.0 s / run
OnePlus 11
Snapdragon 8 Gen 2, 12 GB RAM
Inference speed 1.1 s / run
Xiaomi 13
Snapdragon 8 Gen 2, 12 GB RAM
Inference speed 1.2 s / run
Xiaomi Redmi Note 12 Pro
Dimensity 1080, 8 GB RAM
Inference speed 4.5 s / run (slow path)
Known incompatible
These devices either lack the required Android version or have insufficient memory. The app may install but inference will fail or be impractically slow.
- Android Go edition phones Memory constraints prevent loading the model
- Phones from 2019 or earlier (most) Android version below 12 or RAM below 4 GB
- iPhones / iPads iOS version not yet available — see FAQ
- Wear OS / Galaxy Watch Insufficient memory and missing camera/microphone APIs
About the AI model
Cove uses Google Gemma 4 — Google DeepMind's open-weight on-device model. Two variants are available: E2B (default, ~2-billion effective parameters / 2.5 GB, Q4_0 quantized) covers most scenarios; E4B (optional upgrade, ~4-billion effective parameters / 3.6 GB, Q4_0 quantized) gives higher accuracy on long technical text and rare languages. The model is downloaded once on first launch and shared across all four Cove apps via a shared directory — installing more apps does not duplicate the model. You can also pre-place the model file under /sdcard/GemmaApp/models/ to skip the download. Free tier: 5 AI runs per day per app. Pro (one-time per-app purchase): same model file, unlimited daily runs.
Ready to try Cove?
Free download. If your phone meets the minimum spec, you'll know within 60 seconds if it runs well for you.