Skills
Gemma 4 Ships Under Apache 2.0: On-Device Agent Models with Native Audio/Video Input, 140+ Languages, MoE Architecture Down to 2.3B Effective Parameters
Google DeepMind released Gemma 4 under Apache 2.0 — the most permissive license for a frontier-derived open model family. Four sizes: E2B (2.3B effective), E4B (4.5B effective), 26B A4B MoE (3.8B active), and 31B Dense. All models natively process video and images; E2B/E4B add native audio input. Purpose-built for agentic workflows on-device with 140+ language support. For builders wanting local agents that see, hear, and reason without cloud dependency, Gemma 4 is the most capable option under a fully open license.
Source
↳ Follow the thread