REDDIT · REDDIT// 9d agoMODEL RELEASE

Gemma 4 powers Google’s local AI stack

Google DeepMind launched Gemma 4 as its most capable open model family yet, released under the permissive Apache 2.0 license. The lineup spans edge-focused E2B and E4B models plus larger 26B MoE and 31B dense variants, with support for advanced reasoning, agentic workflows, multimodal input, long context, and offline deployment on phones, laptops, and workstations.

// ANALYSIS

Hot take: this is less a single model drop than Google making a hard push to own the “open but production-ready” lane for local and edge AI.

–Apache 2.0 is the headline feature for builders who want commercial flexibility without license friction.
–The E2B/E4B models matter as much as the bigger ones because they target real on-device and offline use cases, not just benchmark chasing.
–The 26B MoE and 31B dense models give Google a credible workstation-class option for local agents and coding workflows.
–If the reported performance holds up in practice, Gemma 4 is a direct threat to the current default open-weight stacks for local deployment.

// TAGS

gemma-4googledeepmindopen-sourceapache-2.0llmmultimodalon-deviceedge-ailocal-ai

DISCOVERED

9d ago

2026-04-03

PUBLISHED

9d ago

2026-04-03

RELEVANCE

9/ 10

AUTHOR

Much_Ask3471