A hands-on experience report running Google DeepMind's Gemma 4 open-weight model family locally using LM Studio and llama.cpp. The E4B variant stands out for its built-in audio (ASR) support and multimodal capabilities, which larger models in the family lack. The author finds image analysis competitive with cloud AI, appreciates the Apache 2.0 license, and concludes that open-source local models are closing the gap with cloud AI faster than expected.

5m read timeFrom xda-developers.com
Post cover image
Table of contents
Google built one of the most accessible open modelsMy first run with Gemma 4 in LM studioThen I took a detour with llama.cpp
672 Impressions