Explore AI

Gemma

Google · United States

Made by
Google · United States
Weights licence
Apache-2.0
Sizes
Small up to 31B, including Mixture-of-Experts (26B with 4B active)
Notable
gemma-4-31B-it · gemma-4-26B-A4B-it

Where it is strong

Unusually permissive terms for models this capable, and strong at long documents. The Mixture-of-Experts variants give near-30B quality at a fraction of the memory.

Where it is weak

A smaller ecosystem than Llama or Qwen, so fewer community quantisations and fine-tunes to choose from.

Our read

The best choice when you have a long document to work through and limited memory.