Off Grid adds on-device Gemma 4 support
Off Grid, the open-source offline-first mobile AI app, shipped support for Gemma 4 E2B and E4B edge variants on Android, with iOS still coming soon. The update keeps inference fully on-device, combining long-context chat, camera-based vision, Whisper speech-to-text, Stable Diffusion image generation, and tool calling in one app. The maker says it runs on the phone’s NPU/CPU with no server or Python setup and reports roughly 15–30 tok/s on Snapdragon 8 Gen 3 and Apple A17 Pro devices.
This is a strong mobile AI update because it turns Gemma 4 into something people can actually carry and use offline, not just benchmark on a laptop.
- –The main differentiator is practical integration: chat, vision, audio, image generation, and tools in one phone app.
- –The 128K context window matters more on mobile than on desktop because it makes long docs and code workflows feasible without a cloud round-trip.
- –The claim that E2B fits in under 1.5GB RAM is the most impressive part, but real-world performance will still vary a lot by device thermals and memory bandwidth.
- –The launch is credible for the open-source crowd because both the model and app licensing are permissive, but the “open all the way down” framing is still about open weights, not fully open model training data.
DISCOVERED
112d ago
2026-04-09
PUBLISHED
112d ago
2026-04-09
RELEVANCE
AUTHOR
CamusCave