Llama 4

What Llama 4 is for

Llama 4 takes images as well as text. Meta shipped it in 2025-04 with 2 sizes (Scout 17B (109B MoE), Maverick 17B (400B MoE)), 10M of context and a Llama 4 Community 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 vision + long context (10M tokens).

Sizes and memory

Our catalogue does not carry a per-size memory breakdown for Llama 4. What it records instead is a floor of 64 GB and a recommended 96 GB for the default configuration. Until a per-size figure exists, treat those two numbers as the whole guidance rather than reading a size table that would have to be invented.

Hardware that runs Llama 4

Without a per-size memory figure there is no honest way to say which card holds Llama 4. Its recorded floor of 64 GB rules out every consumer card in our hardware guide, which tops out well below that.

What people use it for

The catalogue lists 4 recorded uses for Llama 4: vision, long-context, agentic, multimodal. It is tagged vision, multimodal, MoE, long-context, tools.

Where Llama 4 sits in the Llama family

Our catalogue holds 7 Llama entries. Ordered by release date, Llama 4 (2025-04) is the 7th of them, and the most recent.

The Llama 4 Community License is a vendor community licence rather than a standard open-source one: usable, redistributable with attribution, and carrying conditions that are worth reading before you build a product on it.

Context window: 10M

10M tokens is an unusually large window even in 2026 — on the order of 8 million words. Filling it is a memory problem before it is a quality problem: the KV cache grows linearly with the tokens you actually put in, and the 96 GB recommendation above does not assume you will.

How to run Llama 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 Llama 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 "Llama 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.