← All models

Model

Muse Glimmer 30B

Made by Meta (Superintelligence Labs) · License: Apache 2.0

A free model for chat, code, questions about images and long documents. You start it with one command.

WARNING: Do not use muse-glimmer:30b-mlx on a Mac. It has open bug reports (1, 2).

Will it run on my computer?

Mac

16 GBToo big24 GBToo big32 GBTight36 GBTight48 GBFits64 GBFits96 GBFits128 GBFits192 GBFits256 GBFits512 GBFits

NVIDIA graphics card

8 GBToo big12 GBToo big16 GBToo big24 GBTight32 GBFitsTwo 24 GB cardsFits

AMD Ryzen AI Max

64 GBFits128 GBFits

This is our estimate from the download size. We did not test it on each computer. “Smaller version” means a smaller download than the command below. Tap your computer to see the exact command.

Start it

  1. Install Ollama from ollama.com/download.
  2. Open Terminal.
  3. Paste this command:
ollama run muse-glimmer

What happens next

If it breaks

  1. On a Mac, the -mlx version is slow, or ollama ps shows it stuck on "Stopping...".

    Two open bug reports. One says the -mlx tag is really built from an NVIDIA-only format (NVFP4) and ran about 12 tokens per second on an M3 Max. The other says it gets stuck on "Stopping..." on a 32 GB M4 Air, while the regular GGUF version finished normally.

    On a Mac, start with the regular tag. If a model is stuck, stop it. An Ollama maintainer says related memory fixes are coming in a future release, so keep Ollama updated.

    ollama stop muse-glimmer:30b-mlx

    Source: 1 · 2

  2. Quick answers on hard coding tasks, or slow answers on simple questions.

    Muse Glimmer sets how hard it thinks from a line in the system prompt, not from a switch.

    Inside the chat, set it to high for coding and agent work, or to low for quick chat.

    /set system "Reasoning strength: high"

    Source: 1 · 2

  3. It forgets the start of a long chat, or ignores most of a document you pasted.

    Ollama does not use the model's full memory by default. Its docs say it starts at 4K tokens on graphics cards under 24 GB and 32K on 24-48 GB. When a chat runs past that, the oldest messages are dropped without any warning.

    Raise the context length with the slider in the Ollama app's Settings, or type this in a Terminal chat. It uses more memory, so check the result with ollama ps (the CONTEXT column).

    /set parameter num_ctx 32768

    Source: 1 · 2 · 3

  4. Replies crawl out a word at a time.

    The model did not fit in graphics memory, so Ollama quietly ran part of it on the CPU. Speed drops a lot, but you get no error.

    Run ollama ps while the model runs. Look at the PROCESSOR column. If it does not say 100% GPU, free some memory. Close other models with ollama stop <name>, lower the context length, or pick a smaller version.

    ollama ps

    Source: 1 · 2

  5. You want it to read a screenshot or a chart.

    It takes text and images mixed together, and Meta pitches it for reading screenshots and documents.

    Drag the image into the Ollama app, or put the file path in your message in Terminal.

    ollama run muse-glimmer "What does this screenshot show? ./screen.png"

    Source: 1 · 2

Sizes and versions

Will it run on my computer?

For muse-glimmer (18 GB download). Context up to 128K, but Ollama starts much lower, see below.

8 GB graphics cardWon't fit
16 GB graphics cardWon't fit
24 GB graphics cardTight
32 GB graphics cardFits
16 GB MacWon't fit
24 GB MacWon't fit
32 GB MacTightmuse-glimmer:30b-mlx: open bugs, try only if you want
64 GB+ MacFitsmuse-glimmer:30b-mlx: open bugs, try only if you want
AMD Ryzen AI Max+ PC, 128 GB (graphics memory set to 96 GB)Fits
  • This is our estimate from published download sizes. We did not test these ourselves.
  • Fits = download x 1.25 + 2 GB fits in usable memory. That leaves room for longer chats.
  • Tight = only download + 1.5 GB fits. Short chats work, but long chats or documents may spill onto the CPU and slow down.
  • Usable memory on a graphics card is all of its graphics memory.
  • On a Mac it is 2/3 of memory up to 32 GB and 3/4 at 64 GB+. By default macOS keeps part of memory away from the GPU. Apple says a 32 GB Mac gives the GPU 21 GB and a 64 GB Mac 48 GB.
  • On an AMD Ryzen AI Max+ PC with 128 GB, it is the graphics memory you set in AMD Software. AMD says you can set up to 96 GB.
  • Apple · AMD

Sizes checked 2026-10-19.

START HERE Easy: Ollama

INSTALLOllama. ollama.com/download
RUN
ollama run muse-glimmer
SIZE18 GB download
NEEDSA 24 GB graphics card or a Mac with 32 GB of memory (tight on both).
NOTEIt reads images too: drag a picture into the Ollama app, or put the image's file path in your message in Terminal. For harder coding or agent tasks, tell it how hard to think: /set system "Reasoning strength: high".
ALSOOn a Mac, the MLX build (19 GB) has open bug reports, so start with the command above. Use this only if you want to try it
ollama run muse-glimmer:30b-mlx
SOURCEsource · checked 2026-10-19

Recommended quants

  • Official GGUF KQuant-17GB Q4_K_M (16.8 GB file)
    meta-models/Muse-Glimmer-30B-GGUF
    Meta's 'start here' build for a 24 GB graphics card (~1.0% average degradation per Meta). Add mmproj (1.4 GB) for images and the optional DFlash drafter (1.6 GB) for faster replies.
  • Official GGUF KQuant-Dynamic Q4_K_XL (19.7 GB file)
    meta-models/Muse-Glimmer-30B-GGUF
    Higher-quality build aimed at a 32 GB graphics card (~0.2% average degradation per Meta).
  • Unsloth Dynamic GGUF (UD-Q2_K_XL … UD-Q8_K_XL)
    unsloth/Muse-Glimmer-30B-GGUF
    Unsloth guide: 2-bit 12-14 GB, 4-bit 17 GB+, 6-bit 20-22 GB, 8-bit 34 GB total memory. Full BF16 is about 58 GB.
Scores and how it compares

Independent score: Artificial Analysis Intelligence Index 17 (high), checked 2026-10-19 · source.

AA tested it at high reasoning strength.

Arena (ranks our computer pages):

Checked 2026-10-19.

Pick this if you specifically want Meta's Apache 2.0 model built for tool use and agents.

Skip this if you want the strongest model for a 24 GB card or a 32 GB Mac, where this one is tight. Gemma 4 31B or Qwen3.8 27B score higher there on every Arena board we use.

Compare: Qwen3.8-27B · Gemma 4 (26B-A4B / 31B) · Pick your computer

BEST PICK ONNot the top pick on any computer page right now.
Settings and apps

Chat template Works

Official chat_template.jinja ships with the weights. Reasoning strength (low / medium / high / xhigh) is set with a line in the system prompt, not a flag.

Fix: Add the system line "Reasoning strength: high" for coding and agent work, lower for quick chat. In the Ollama chat window: /set system "Reasoning strength: medium".

Sampling settings

Officialtemp 1.0, top_p 0.95, top_k 64
Reasoning strengthlow / medium / high / xhigh, set with "Reasoning strength: <value>" in the system prompt
Context131,072 tokens (Unsloth: up to 262,144)

Runtimes

OllamaWorksOfficial library muse-glimmer: :30b / latest (18 GB, text + image) and :30b-mlx (19 GB) for Apple Silicon, plus DFlash tags, as of Oct 19.
llama.cppWorksNeeds build b10353 or newer (support merged Aug 10). Older builds refuse to load it; if replies start with to=self<|message|>, update.
LM StudioWorksListed as meta/muse-glimmer; LM Studio says the smallest build needs at least 26 GB of memory.
MLXWorksRuns with Ollama's MLX tags (:30b-mlx), which Ollama says support DFlash and image input.
vLLMUnverifiedNot checked for this card.

Tool calling in agent tools

Claude Code (local endpoint)UnverifiedOllama documents ollama launch claude --model muse-glimmer.
OpenCodeUnverifiedOllama documents ollama launch opencode --model muse-glimmer.
Sources
MADE BYMeta (Superintelligence Labs)
LICENSEApache 2.0 · model page · checked 2026-10-19
RELEASED2026-08 (weights on Hugging Face Aug 9; Meta paper published Oct 19)
CHECKED2026-10-19 · 6 sources

Meta's 30B open model for chat, images and agents (Apache 2.0). One Ollama command, about 18 GB, tight on a 24 GB graphics card or a 32 GB Mac.

Known issues

Sources checked

Spotted something wrong or a new fix? Email corrections to [email protected]. Corrections are reviewed before publishing.