Gemma 4

What Gemma 4 is for

Gemma 4 takes images as well as text. Google shipped it in 2026-05 with 5 sizes (e2b, e4b, 12b, 26b, 31b), 128K of context and a Gemma License licence. Running a vision-language model locally costs more memory than a text model of the same parameter count, because the image encoder and the projected image tokens both occupy the context. The catalogue records its strength as frontier-class multimodal with strong reasoning.

Sizes and memory

The 12b needs about 8 GB of weights at Q4_K_M and the 31b about 20 GB, a spread of 12 GB across the family. The catalogue's default pick is 26b. Recommended VRAM adds roughly half again on top of the weights for the KV cache and the runtime.

Size	Weights at Q4_K_M	Recommended VRAM	Smallest card in our guide that holds it
12b	8 GB	12 GB	NVIDIA RTX 3060 12GB
26b	17 GB	26 GB	NVIDIA RTX 3090
31b	20 GB	30 GB	NVIDIA RTX 3090

Hardware that runs Gemma 4

18 of the 19 cards in our hardware guide hold at least one size of Gemma 4 at Q4_K_M. The cheapest that does is the NVIDIA RTX 3060 12GB at about $250, running the 12b.

The largest size any of them holds is the 31b at 20 GB, on the NVIDIA RTX 3090 (24 GB at 936 GB/s). Bandwidth, not capacity, sets the speed once a size fits, so a card further down the list with faster memory will generate more tokens per second on the same weights than a larger, slower one.

What people use it for

The catalogue lists 5 recorded uses for Gemma 4: frontier, vision, audio, tools, agentic. It is tagged vision, tools, thinking, audio, frontier.

Where Gemma 4 sits in the Gemma family

Our catalogue holds 3 Gemma entries. Ordered by release date, Gemma 4 (2026-05) is the 3rd of them, and the most recent.

It ships under the Gemma License, which is neither a standard open-source licence nor a closed one. Read the terms before shipping anything commercial on top of it.

Context window: 128K

128K tokens is roughly 96,000 words at the usual ratio, which is a small repository or a whole book. Two cautions come with a window this size: the KV cache at full length can rival the weights for memory, and models rarely use the far end of their advertised context as well as they use the near end. Treat it as headroom, not as a promise.

How to run Gemma 4 locally

Our catalogue does not record registry tags, so look the current tag up in the Ollama library or on Hugging Face before pasting these. The commands below are the shape of the workflow, not a copy-and-paste recipe.

Option 1: Ollama (simplest)

# find the tag for Gemma 4 at ollama.com/library
ollama run <tag>

Option 2: Mullama (production)

mullama pull <tag>
mullama run <tag>

Option 3: llama.cpp (CLI)

# download a GGUF from Hugging Face, searching for "Gemma 4 GGUF"
./llama-cli -m model.Q4_K_M.gguf -p "Hello, AI!"

Option 4: Python with Mullama or llama-cpp-python

from mullama import Model, Context
model = Model.load("model.Q4_K_M.gguf", n_gpu_layers=99)
ctx = Context(model, n_ctx=4096)
print(ctx.generate("Hello, AI!", 256))

Sources

Model record from src/data/models.json (2026-06-29); card capacities from src/data/gpus.json (2026-06-29). Upstream: Curated from ollama.com/library + community benchmarks + paperswithcode + huggingface.