GoogleGemma 4
Gemma 4 26B A4B
Gemma 4's mixture-of-experts checkpoint.
What the release provides
The card lists 25.2B total and 3.8B active parameters, text/image input, 256K context and Apache 2.0 licensing.Google Gemma 4 model card
How to use it well
Practical guidance. Plan weight storage from the total model size. Then measure token throughput with the selected serving engine and batch size. Sparse activation reduces computation per token, but it does not mean the full deployment has the memory footprint of a small dense model.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | google/gemma-4-26B-A4B-it |