Meta releases Muse Glimmer, 30B open-weight model for local agents

devAug 10, 20263875

Meta Platforms released Muse Glimmer, a 30-billion-parameter open-weights model published under an Apache 2.0 license and intended to run locally on consumer hardware such as a single laptop GPU. Meta positions Muse Glimmer as purpose-built for agentic workflows with planning, tool use, self-correction, and a dedicated perception encoder to support multimodal inputs. Meta and third-party publishers say fourth-bit quantized builds can fit under about 20, 24 GB of VRAM, enabling local inference on many single-GPU setups. The company also signaled it will publish an open-weights version of its larger Muse Spark 1.2 model in the coming weeks, a detail reported by the Wall Street Journal. Reuters noted CEO Mark Zuckerberg used the launch to urge lower U.S. barriers for open-source, open-weight models. Meta has uploaded Muse Glimmer weights and community runtimes and guides are appearing on Hugging Face and third-party repositories to support local use and experimentation.

amos

@fasterthanli.me

> Exclusive: Muse Code Sends Codex and Claude Instructions to Meta by Default Meta? Cavalier about privacy boundaries?? That'd be a first runtimewire.com/article/excl...

27Yesterday

Dabo4

i think this makes sense? if I know one thing about it, Muse Code is a tool to get an extremely, extremely cheap model in exchange for giving data to Meta. & of course it sends the imported system prompt files on first run, they're system prompt files

Reid Marlow1

Yeah, importing foreign personal rules outside the selected workspace should be opt-in. Cross-tool convenience is not worth making the privacy boundary depend on someone noticing a startup banner.

Unsloth AI

@unsloth.ai

Meta releases Muse Glimmer, a new 30B open model that runs on 18GB RAM. Apache 2.0 licensed and built for agentic coding, it supports vision and is the strongest agentic model for its size. Run and train the model via Unsloth. GGUF: huggingface.co/unsloth/Muse... Guide: unsloth.ai/docs/models/...

134h ago

Techmeme

@techmeme.com

Meta releases Muse Glimmer, a new open-weight model, and will release an open-weight version of its most advanced model, Muse Spark 1.2, in the coming weeks (Meghan Bobrowsky/Wall Street Journal) Main Link | Techmeme Permalink

64h ago

Pekka Lund

@pekka.bsky.social

Turns out Meta was serious in their support of open source AI. Muse Spark 1.2 will be the leading open US model if they release it soon.

Mark Zuckerberg @finkd

Today we're also opening the weights for Muse Glimmer, a great 30B parameter dense model that can run locally. Soon we'll also release the weights for Muse Spark 1.2, our latest foundation model. Meta is a strong supporter of open source and I'm proud of these releases. Congrats to @alexandr_wang and the MSL team for all your great work on these models.

94h ago

Stuart Gray

@sgray.bsky.social

Meta have just dropped Muse Glimmer 30B. As a dense model it’s a bit of a mixed bag with benchmarks suggesting it’s broadly comparable with Qwen 3.6 27B and Gemma 4 31B, and comes with control over reasoning effort. huggingface.co/meta-models/...

@holeafuk.bsky.social

Can you use Meta's new AI tools to quickly prototype a game? Of course you can. Should you? Maybe not just yet. This weeks blog is about testing Meta's AI, Muse: by building LLM Wars. holeaf.uk/blog/testing...

Yesterday

Poster | Crypto News

@cryptonews-poster.bsky.social

Meta open-sources Muse Glimmer (30B params), an agent model capable of running on 24GB VRAM (4-bit quantized)! Features planning, tool use, self-correction. Weights on Hugging Face, with inference support coming soon to Ollama, LM Studio, and more! #crypto #blockchain #news

14h ago

Hacker News Top 100 Bot (Unofficial)

@hn100.bsky.social

Meta Muse Glimmer – open weights 30B local coding model Discussion

14h ago

ClawBeat.co

@clawbeat.bsky.social

Meta reasserts its open-source AI leadership with Muse Glimmer, a new local, agentic, multimodal model poised to democratize advanced AI development.

13h ago

Knut Jägersberg

@knut-jagersberg.wsocial.eu

unsloth/Muse-Glimmer-30B-GGUF Authors: Meta Superintelligence Lab Muse Glimmer is a 30-billion-parameter causal language model with a dedicated perception encoder, distilled from Muse Spark and purpose-built for autonomous agentic tasks on consumer hardware. huggingface.co/unsloth/Muse...

13h ago

coridauber.bsky.social

@coridauber.bsky.social

Meta Unveils ‘Open Source’ Version of Its Most Powerful A.I. Model www.nytimes.com/2026/08/10/t...

12h ago

Techimo

@techimo.bsky.social

Meta's new AI model, Muse Glimmer, is here! This 30B parameter distilled version of Muse Spark 1.2 runs on a single GPU, perfect for intelligent agent tasks like scheduling & file org. Meta CEO Mark Zuckerberg also called for policy review to maintain US leadership in open-weight models.

13h ago

CNA - Channel News Asia (Unofficial)

@cna-poster.bsky.social

Meta launches new AI model as Zuckerberg champions open-weight push - https://www.channelnewsasia.com/business/meta-launches-new-ai-model-zuckerberg-champions-open-weight-push-6309576

14h ago

Firethering

@firethering.com

Zuckerberg Wrote 14 Pages About Open AI. His Best AI Model Is Still Closed #openAI #tech #AI firethering.com/meta-muse-gl...

13h ago

3 sources

leverageonheroesmedia.comMeta Releases Muse Glimmer AI Model People Can Run on Their Laptop Meta Platfor techmeme.comMeta releases Muse Glimmer, a new open-weight model, and will release an open-weight version of its most advanced model, Muse Spark 1.2, in the coming weeks (Meghan Bobrowsky/Wall Street Journal)upstract.comMeta launches new Muse Glimmer AI model

Meta releases Muse Glimmer, 30B open-weight model for local agentsdev

Meta releases Muse Glimmer, 30B open-weight model for local agents

Related