REDDIT · REDDIT// 7d agoMODEL RELEASE

Gemma 4 26B MoE hits Apache 2.0

Google's Gemma 4 family debuts with a 26B MoE model that rivals 30B+ dense models while running at 4B speeds on consumer hardware. The release marks a major shift for the open-weights ecosystem with the adoption of the Apache 2.0 license.

// ANALYSIS

Gemma 4's transition to a Mixture-of-Experts architecture and the Apache 2.0 license effectively ends the debate for local LLM users. The MoE architecture delivers a 2-3x inference speedup over dense models by activating only 3.8B parameters per token. Native reasoning and a 256K context window enable complex analysis, while the new license removes commercial restrictions that previously hampered production adoption.

// TAGS

gemma-4llmopen-weightsmoebenchmarkopen-source

DISCOVERED

7d ago

2026-04-05

PUBLISHED

7d ago

2026-04-04

RELEVANCE

10/ 10

AUTHOR

simracerman