Gemma 4 E4B

Good fit for

People evaluating an edge-oriented multimodal model with configurable thinking.

Why consider it

The exact Google card documents image and audio input, text output, coding and a 128K context window for E4B.

Main limitations

E4B means effective parameters, not total storage or a 4GB memory guarantee. Audio transcription is not full Voice Chat. Actual supported modalities depend on runtime and client. No benchmark score is presented as hands-on validation.

Cost and access

No subscription or per-token hosted fee for local inference within permitted uses. Hardware, electricity and setup are separate. Apache 2.0 is confirmed in both the runtime package and Google’s exact model card.

Getting started

Install a current compatible Ollama version, review the exact license and run: ollama run gemma4:e4b. API refers to the local runtime endpoint, not a hosted API included with the weights.