Google Explains Why Android AICore Temporarily Spikes Storage During Model Updates
Key Takeaways
- ▸Android AICore temporarily stores both old and new Gemini Nano model versions for up to 3 days during updates as a stability failsafe
- ▸The dual-version approach allows instant rollback if updates fail, preventing users from needing to re-download gigabytes of model data
- ▸AICore supports six major on-device AI features: advanced proofreading, automatic speech recognition, scam detection, smart replies, summarization, and translation
Summary
Google has published detailed documentation explaining why its Android AICore feature occasionally experiences temporary storage increases. The platform runs Gemini Nano models directly on-device to power features like advanced proofreading, speech recognition, scam detection, smart replies, summarization, and translation. During model updates, AICore intentionally keeps both the old and new model versions for up to three days as a failsafe mechanism. If the new update encounters errors, the device can instantly revert to the previous version without requiring users to re-download gigabytes of data. Once the system confirms update stability, the extra storage space automatically clears. Google emphasizes that on-device AI processing provides significant benefits including enhanced privacy (sensitive information never leaves the device), offline functionality, and faster performance compared to cloud-based alternatives.
- On-device processing delivers privacy advantages, offline functionality, and reduced latency compared to cloud-based AI services
