Meta Muse models

Meta's newer open models: Muse Glimmer, a 30B Apache 2.0 model for local agents, and Muse Spark Open, an 816B multimodal MoE with a 1M-token context.

Open weights
Models 2
Context up to 128K
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 Meta Muse
Guides
FAQ
Compare

At a glance

Type: Open-weight models
Developer: Meta (United States)
First release: Aug 2026
Models tracked: 2 (2 current)
Longest context: 128K tokens
Abilities: Tool callingReasoning modeImage input
Licence: Muse Glimmer is Apache 2.0. Muse Spark Open uses the Muse Model License, which has limited commercial terms; read it before you build a product on it.
Smallest local option: Muse Glimmer 30B · ~20.5 GB
Website: ai.meta.com
Hugging Face: huggingface.co/meta-models

Newest model

Muse Spark Open

Released Sep 2026

Smallest option

Muse Glimmer 30B

~20.5 GB VRAM · 18 GB download

Meta Muse models and versions

Model	Released	Sizes	Context	Abilities	Licence	Status
Muse Spark Open Frontier-size multimodal model for coding and long agent runs	Sep 2026	816B 816B total, 42B active (MoE) Too large for most servers · Ollama cloud	—	ToolsReasoningVision	Muse Model License	Current
Muse Glimmer Local agent model with tools, reasoning and image input	Aug 2026	30B	128K	ToolsReasoningVision	Apache 2.0	Current

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 Meta Muse 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	Command
Muse Glimmer 30B	18 GB	31 GB	57 GB	~20.5 GB	24 GB 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 Muse Glimmer locally

# Install Ollama (Linux), then pull and chat
curl -fsSL https://ollama.com/install.sh | sh
ollama run muse-glimmer:30b

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": "muse-glimmer:30b", "messages": [{"role": "user", "content": "Hello"}]}'

# Full-precision weights from Hugging Face, OpenAI-compatible API on :8000
vllm serve meta-models/Muse-Glimmer-30B-assistant

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 Meta Muse is good for

Always-on local agents

Muse Glimmer was distilled from Muse Spark for multi-step tool use on consumer hardware.

Private multimodal assistant

Glimmer takes images and runs offline on a 24 GB GPU or a Mac with 32 GB.

Large-scale coding agents

Muse Spark Open targets long-horizon software engineering on data-centre GPUs.

Strengths and limits

Strengths

Glimmer is Apache 2.0, unlike the Llama models.
Built for agents: tool use, failure recovery and reasoning in one model.
Spark Open has a 1M-token context.

Watch out for

New family with a small fine-tune and tutorial ecosystem so far.
Spark Open needs data-centre hardware and has a restrictive licence.

About Meta Muse

Muse is the name Meta uses for its newest models. Muse Glimmer (August 2026) is a 30B model with a vision encoder, distilled from Meta’s larger Muse Spark and built to run autonomous agent tasks on consumer hardware without a network connection. It is released under Apache 2.0, a more permissive licence than any Llama model.

Muse Spark Open (September 2026) is a natively multimodal mixture-of-experts model with 816B parameters (42B active) and a 1M-token context window, aimed at coding and long agent runs. It uses the Muse Model License, which allows free use for many developers but limits some commercial use. On Ollama it is offered through the cloud tier because it does not fit on ordinary servers.

Guides for running Meta Muse

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 llama.cpp on Ubuntu 26.04 (CPU and NVIDIA GPU) 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 Meta Muse model?

The newest Meta Muse model is Muse Spark Open, released in Sep 2026.

How much VRAM do I need to run Meta Muse locally?

The smallest option, Muse Glimmer 30B, is a 18 GB download at 4-bit and needs about 20.5 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 Meta Muse with Ollama?

Install Ollama, then run "ollama run muse-glimmer:30b". 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 Meta Muse commercially?

Muse Glimmer is Apache 2.0. Muse Spark Open uses the Muse Model License, which has limited commercial terms; read it before you build a product on it. Always read the licence of the exact model you deploy; it is linked in the models table.

Compare Meta Muse

Muse Glimmer vs Qwen3.8

← All AI models