Gemma 4 E2B/E4B ships as production-ready on-device AI for Android — Apache 2.0, multimodal, per-layer embeddings

Google's Gemma 4 family — E2B, E4B, 26B A4B MoE, 31B Dense — launched in April with E2B and E4B specifically targeted at on-device Android and laptop deployment. All Gemma 4 models accept text and image input and analyze video as frame sequences; E2B and E4B additionally support audio input. Per-layer embeddings improve parameter efficiency for on-device contexts. The launch is the cleanest 'on-device AI is production-ready' signal of 2026 H1.

The license-and-modality combination is the structural differentiator. Apache 2.0 + multimodal text/image/video/audio + sub-billion-parameter footprint clears the procurement bar for any Android OEM or laptop vendor that wants on-device AI without revenue-share licensing or cloud round-trips. The 'Goldilocks zone' for on-device language models the industry has been chasing for two years now has a concrete product reference.

For Phi-4's 14B / 5.1 GB peak memory footprint, the competitive picture sharpens. Phi-4 reasons well above its weight class but leaves only 2.9 GB for the rest of the OS — a deployment ceiling that constrains Phi to higher-spec edge devices. Gemma 4 E2B/E4B sits inside the deployment envelope of mid-tier Android phones. The market split: Phi for premium edge, Gemma for mainstream edge.

Edge AI Vision — Gemma 4 edge → · Google AI — Gemma 4 model overview → · MindStudio — Gemma 4 edge deployment →