gemma-4-26B-A4B-it

gemma-4-26B-A4B-it

Instruction-tuned Gemma 4 model from Google that reads text and images and writes text, with configurable thinking, function calling and a context window of up to 256K tokens. For developers and researchers.

Made by google on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: gemma-4-26B-A4B-it is the instruction-tuned 26B A4B size of Gemma 4, an open-weights family from Google DeepMind. It accepts text and images, writes text, and supports over 140 languages.

Gemma 4 comes in dense and mixture-of-experts designs and in five sizes. The family offers configurable thinking modes, native function calling and a system role for more controlled chats.

Who it suits: Developers and researchers who want an open model for coding, reasoning, agents and image questions, on consumer GPUs or workstations.

Industry | Who uses it | How they use it | Example
Software | Developers of coding tools and agents | Use function calling and reasoning modes | Build an agent that calls your own functions
Product teams | Builders of multilingual apps | Serve users in many languages | Answer support questions in one of 140 plus languages
Data teams | Teams working with documents and images | Ask questions about pictures and text | Describe a chart in an uploaded image
Research | Open model researchers | Fine-tune or evaluate an open model | Adapt it to a domain

Skip it if you need audio input, since the card gives audio to the smaller E2B and E4B models and to the 12B model.

Where to use it:

• Consumer GPUs and workstations, which the card names as the target for this size
• Windows, macOS or Linux with Python, to download the files
• Programs that use the Transformers library

Use case | Who | What they get back
Coding and reasoning | Developers | Text answers with optional thinking
Agent workflows | Agent builders | Function calls from the model
Image questions | App builders | Text answers about a picture you supply

How to set it up:

• Python and pip
• A GPU or workstation with enough memory, since the weights are about 51.6 GB in BF16 and about 12.9 GB at 4-bit by estimate
• The Transformers library to load the model

1. Install the Hugging Face tools.
2. Download the model files.
3. Load the model in your own code and follow the usage notes on the model page.

pip install -U huggingface_hub
hf download google/gemma-4-26B-A4B-it

How to use it: Send text, or text with an image, through the chat template and read back text. You can set a system role, and the card says thinking modes are configurable.

Example: You send a picture of a chart together with the question of what trend it shows. The model replies in text with its reading of the chart.

For newcomers and experts: If you are new to open models, start with a short text question and no image, then add a picture. Use a quantized copy if your graphics card has limited memory.

If you know the field, the family mixes local sliding-window attention with global attention and always makes the last layer global. Global layers use unified keys and values and p-RoPE to save memory on long inputs.

Text or image input
Chat template with optional system role
Hybrid local and global attention
Text output

Good to know:

• Hugging Face lists 424 quantized and 241 fine-tuned versions made by other people
• Five sizes make up the family, running from the small E2B up to the 31B
• The small models have a 128K context window and the medium models reach 256K

Title: Context length of the dense Gemma 4 sizes in thousands of tokens
E2B: 128 K tokens
E4B: 128 K tokens
12B Unified: 256 K tokens
31B Dense: 256 K tokens

• License: Apache 2.0
• Cost to run: none named
• Data it reaches: the text and images you send it
• Not for: audio input, which the card assigns to the E2B, E4B and 12B models

On Hugging Face, 13 November 2026

• Task: image text to text
• Library: transformers
• Parameters: 25.8B
• Weights: about 51.6 GB in BF16; about 12.9 GB at 4-bit (estimate)
• Formats: Safetensors
• Downloads in the last 30 days: 5,304,019
• Likes: 1,758
• License: apache-2.0
• Based on: google/gemma-4-26B-A4B
• Quantized versions: 430
• Fine-tunes: 244
• Tags: transformers, safetensors, gemma4, image-text-to-text, conversational

Links

• Model: https://huggingface.co/google/gemma-4-26B-A4B-it
• Files: https://huggingface.co/google/gemma-4-26B-A4B-it/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download google/gemma-4-26B-A4B-it

Then follow the usage in its model card. Nothing in this kit runs by itself.