Llama models

Meta's open-weight model family. Llama 3.1 is still the most-pulled model on Ollama; Llama 4 added mixture-of-experts and up to a 10M-token context.

Open weights
Models 6
Context up to 10M
Updated Nov 2026

Hugging Face Compare

On this page

At a glance
Models & versions
VRAM & download size
Run it locally
Best uses
Strengths & limits
About Llama
Guides
FAQ
Compare

At a glance

Type: Open-weight models
Developer: Meta (United States)
First release: Feb 2023
Models tracked: 6 (1 current)
Longest context: 10M tokens
Abilities: Tool callingImage input
Licence: Llama Community License: free for commercial use, but companies with more than 700 million monthly users need a separate licence from Meta, and you must show "Built with Llama".
Smallest local option: Llama 3.2 1B · ~1.5 GB
Website: llama.com
Hugging Face: huggingface.co/meta-llama

Newest model

Llama 4

Released Apr 2025 · 10M context

Smallest option

Llama 3.2 1B

~1.5 GB VRAM · 808 MB download

Llama models and versions

Model	Released	Sizes	Context	Abilities	Licence	Status
Llama 4 Multimodal MoE; Scout has a 10M-token context	Apr 2025	Scout 109B · Maverick 400B 109B (Scout), 400B (Maverick) total, 17B active (MoE)	10M	ToolsVision	Llama 4 Community License	Current
Llama 3.3 70B model close to Llama 3.1 405B quality	Dec 2024	70B	128K	Tools	Llama 3.3 Community License	Previous generation
Llama 3.2 Vision Image reasoning	Sep 2024	11B · 90B 11B, 90B	128K	Vision	Llama 3.2 Community License	Previous generation
Llama 3.2 Tiny models for edge devices and CPUs	Sep 2024	1B · 3B 1B, 3B	128K	Tools	Llama 3.2 Community License	Previous generation
Llama 3.1 Most-pulled model on Ollama; reliable 8B all-rounder	Jul 2024	8B · 70B · 405B 8B, 70B, 405B	128K	Tools	Llama 3.1 Community License	Previous generation
Llama 3 Only 8K context; use 3.1 instead	Apr 2024	8B · 70B 8B, 70B	8K	Text only	Meta Llama 3 License	Legacy

Sizes, context windows and abilities come from the Ollama library and Hugging Face model cards, checked 14 Nov 2026. Release months are the first public upload. Spotted a mistake? Tell us.

How much VRAM does Llama need?

Download sizes per Ollama build and the GPU memory each needs with an 8K context. On a CPU-only server the same number is the RAM you need. Longer contexts need more memory; work it out with the LLM VRAM calculator.

Model	4-bit (default)	8-bit	16-bit	VRAM (4-bit)	Fits on
Llama 4 Scout 109B	67 GB	—	—	~74.5 GB	80 GB GPU
Llama 4 Maverick 400B	245 GB	—	—	~270 GB	Multi-GPU
Llama 3.3 70B	43 GB	75 GB	141 GB	~48 GB	48 GB GPU
Llama 3.2 Vision 11B	7.8 GB	12 GB	21 GB	~9.5 GB	12 GB GPU
Llama 3.2 Vision 90B	55 GB	95 GB	177 GB	~61.5 GB	80 GB GPU
Llama 3.2 1B	808 MB	1.3 GB	2.5 GB	~1.5 GB	8 GB GPU
Llama 3.2 3B	2 GB	3.4 GB	6.4 GB	~3 GB	8 GB GPU
Llama 3.1 8B	4.9 GB	8.5 GB	16 GB	~6 GB	8 GB GPU
Llama 3.1 70B	43 GB	75 GB	141 GB	~48 GB	48 GB GPU
Llama 3.1 405B	243 GB	431 GB	812 GB	~268 GB	Multi-GPU

VRAM = 4-bit download + about 10% runtime overhead + 0.5 GB for an 8K context, rounded up. Mixture-of-experts (MoE) models still need all their weights in memory, but they run much faster than a dense model of the same size, so they also work well split between GPU and system RAM.

Run Llama 4 locally

# Install Ollama (Linux), then pull and chat
curl -fsSL https://ollama.com/install.sh | sh
ollama run llama4:scout

New to Ollama? Follow Install Ollama on Ubuntu 26.04 (or 24.04) for GPU drivers, the systemd service and remote access.

# Ollama serves an OpenAI-compatible API on port 11434
curl http://localhost:11434/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"model": "llama4:scout", "messages": [{"role": "user", "content": "Hello"}]}'

# Full-precision weights from Hugging Face, OpenAI-compatible API on :8000
vllm serve meta-llama/Llama-4-Scout-17B-16E-Instruct

vLLM loads the original 16-bit weights, so it needs roughly the 16-bit size above in GPU memory, plus room for the context. See Install vLLM on Ubuntu 26.04.

What Llama is good for

General chat on modest hardware

Llama 3.1 8B and Llama 3.2 3B are well tested and supported by every tool.

Long documents

Llama 4 Scout accepts up to 10M tokens, but the memory for that context is far beyond one GPU.

Fine-tuning

The Llama 3.x models have the largest ecosystem of fine-tunes, adapters and tutorials.

Strengths and limits

Strengths

Huge ecosystem: every runtime, fine-tuning tool and tutorial supports Llama first.
Llama 3.x models are predictable and well understood in production.
Llama 4 Scout has the longest context of any open model (10M tokens).

Watch out for

Licence is not open source: user cap, attribution and an acceptable-use policy.
Llama 4 needs at least 67 GB even at 4-bit; there is no small Llama 4.
Newer families (Qwen3.5, Gemma 4) beat Llama 3.x at the same size.
Meta's newest open models ship under the Muse name, not Llama.

About Llama

Llama is the model family that started the open-weight wave. The original LLaMA (February 2023) was released for research; Llama 2 (July 2023) and Llama 3 (April 2024) came with a commercial licence, and Llama 3.1 (July 2024) added a 128K context and the 405B model. Llama 3.1 8B is still the most downloaded model in the Ollama library.

Llama 3.2 (September 2024) brought tiny 1B/3B models and the first vision models, and Llama 3.3 (December 2024) gave 405B-class quality in a 70B model. Llama 4 (April 2025) switched to mixture-of-experts with native image input: Scout (109B total, 17B active, 10M-token context) and Maverick (400B total, 17B active).

In 2026 Meta started releasing its open models under a new name, Muse. Llama 3.x remains a safe, well-supported choice, but for new projects compare it with Qwen3.5 and Gemma 4, which usually give better quality at the same size.

Guides for running Llama

AI & LLMs How to Install llama.cpp on Ubuntu 24.04 (CPU and NVIDIA GPU) Updated Nov 2026 AI & LLMs How to Install llama.cpp on Ubuntu 26.04 (CPU and NVIDIA GPU) Updated Nov 2026 Tool LLM VRAM Calculator: How Much GPU Memory Do You Need? Updated Nov 2026 AI & LLMs How to Install Ollama on Ubuntu 26.04 and Run LLMs Locally Updated Nov 2026 AI & LLMs How to Install vLLM on Ubuntu 26.04 (OpenAI-Compatible LLM Server) Updated Nov 2026 AI & LLMs How to Install Open WebUI on Ubuntu 26.04 (Self-Hosted ChatGPT Alternative) Updated Nov 2026

Frequently asked questions

What is the latest Llama model?

The newest Llama model is Llama 4, released in Apr 2025. It has a 10M-token context window and supports tool calling, image input.

How much VRAM do I need to run Llama locally?

The smallest option, Llama 3.2 1B, is a 808 MB download at 4-bit and needs about 1.5 GB of GPU memory (or RAM on a CPU-only machine) with an 8K context. Larger sizes need more; see the hardware table on this page, and use the VRAM calculator for longer contexts.

How do I run Llama with Ollama?

Install Ollama, then run "ollama run llama3.2:1b". Ollama downloads the 4-bit build and starts a chat; the same model is also available on the local API at http://localhost:11434.

Can I use Llama commercially?

Llama Community License: free for commercial use, but companies with more than 700 million monthly users need a separate licence from Meta, and you must show "Built with Llama". Always read the licence of the exact model you deploy; it is linked in the models table.

Is Llama open source?

Meta calls it open, but the Llama licence is a custom licence with a 700M monthly-user limit and usage rules, so it does not meet the Open Source Definition. The weights are free to download and use commercially for most companies.

Is there a Llama 5?

No. Meta's newer open models are released under the Muse name (Muse Glimmer and Muse Spark Open).

Compare Llama

← All AI models