Meta Open-Sources 30B Muse Glimmer for On-Device Agents

Meta's Superintelligence Lab released Muse Glimmer, a 30B-parameter open-source multimodal model optimized for on-device agent workflows on consumer hardware. Under Apache 2.0, it supports 128K+ context. Developer tests show it runs efficiently on minimal resources and demonstrates autonomous deployment, environment adaptation, and long-horizon tool use, signaling a shift from passive generation to active system task execution.

Confirmed

• Release and ecosystem support: Meta released Muse Glimmer, a 30B dense multimodal model with Gemma-like architecture and GQA. Hugging Face and UnslothAI provided day-zero support; Unsloth uploaded GGUF quantized versions and deployment guides. Meta also teased upcoming weights for Muse Spark 1.2.
• Autonomous deployment and system optimization: Developer benburtenshaw tested tool-calling abilities; the model self-quantized and deployed to a protected Hugging Face Inference Endpoint, auto-completing health checks and chat completion validation. In advanced tests, it optimized deployment configs for H100 architecture.
• Low-resource tool calling: UnslothAI demonstrated 100+ consecutive tool calls using 2-bit GGUF quantization, consuming only 14GB RAM.
• Local run and experience: Developer MaziyarPanahi ran it on Mac Studio at 30 tok/s API speed; benburtenshaw set up a research agent on Hugging Face Hub.
• Performance and benchmarks: On 24 benchmarks, Muse Glimmer achieved best scores in 12, outperforming Gemma4-31B and Qwen3.6-27B (per kimmonismus). ForsookComparison found its chain-of-thought highly efficient, surpassing popular models of similar size in some scenarios.
• Hardware requirements and quantization: Official says 24GB VRAM needed (m16); danielhanchen retweeted 18GB VRAM. coder543 tested on RTX 3090 with Q4KXL, DFlash, and mmproj supporting 256k context at 22GB VRAM; ydnar shared optimized configs for RTX 3090.
• Coding and inference speed: Ok-Shower7286 achieved 280 t/s on local coding tasks with DFlash and llama.cpp; curiousily on M5 Pro with Q4 used 20GB RAM at 17 token/s.

Unconfirmed

• Whether the model fully supports 256k context on 24GB VRAM: coder543's test shows feasible, but performance may vary with different quantizations.
• Relative capability vs. Qwen: curiousily thinks coding is inferior to Qwen, but kimmonismus cites benchmarks showing superiority over Qwen3.6-27B; disagreement exists.

Why it matters

• These tests demonstrate AI models transitioning from passive text generation to active complex system tasks. The model's autonomous deployment, environment adaptation, and hardware optimization provide valuable reference for future AI-driven ops.
• The ability to run 100+ tool calls on 14GB RAM significantly lowers hardware barriers for on-device complex agents, making persistent local agents feasible.
• Zuckerberg reiterated Meta's commitment to open-source AI, calling for unrestricted model distillation. eyishazyer notes this reflects Meta's dual strategy: keeping flagship models closed for commercialization while open-sourcing distilled compressed versions.

2026-08-10 ~ 2026-08-12 · 108 related posts

Full story(8 episodes)→

• Episode 1: Meta Open-Sources 30B Muse Glimmer for On-Device Agents(2026-08-10, 108 posts)
• Episode 2: Meta to Open-Source 30B Muse Glimmer, Tops Open-Source Benchmarks(2026-08-10, 6 posts)
• Episode 3: Leaked Architectures of New 30B to 400B Parameter Models Spark Discussion(2026-08-10, 2 posts)
• Episode 4: llama.cpp Adds Day-0 Support for Meta's Muse Glimmer(2026-08-10, 2 posts)
• Episode 5: Meta's Open-Source Muse Glimmer 30B Impresses in On-Device Agent Tests(2026-08-11, 14 posts)
• Episode 6: Muse Glimmer Tested: Reaches 92% of Qwen's Intelligence with 1/3 Compute(2026-08-11, 2 posts)
• Episode 7: Meta Releases Local Open-Source LLM as OpenAI Unveils Unrestricted Safety Model(2026-08-12, 2 posts)
• Episode 8: Single GH200 Hits 3323 tok/s in Muse Glimmer 30B Inference(2026-08-12, 3 posts)

• Meta Releases Muse Glimmer 30B Model Weights, Runs on a Single Consumer GPU — alexandr_wang · 2026-08-10
• Meta Releases Muse Glimmer: A 30B Open-Weight Agentic Model Running on 24GB VRAM — alexandr_wang · 2026-08-10
• [source] Meta Releases Muse Glimmer: A 30B Open-Weight Model for Local Agents — AIatMeta · 2026-08-10
• [source] Meta Releases Muse Glimmer: A 30B Open-Weight Model for Local Agents — AIatMeta · 2026-08-10
• Testing Meta's Muse Glimmer: Getting the 30B Model to Quantize Itself — ben_burtenshaw · 2026-08-10
• Deploying Itself: Muse Glimmer Agent Demo — ben_burtenshaw · 2026-08-10
• Research Agent for Muse Glimmer Live on Hugging Face Hub — ben_burtenshaw · 2026-08-10
• Getting AI to Optimize Its Own Deployment for H100: Muse Glimmer Demo — ben_burtenshaw · 2026-08-10
• Meta to Open-Source Muse Glimmer: A 30B Agentic Model Requiring Only 24GB VRAM — huggingface · 2026-08-10
• ByteDance Open-Sources Muse Glimmer: A 30B Model Running on a Single Consumer GPU — MaziyarPanahi · 2026-08-10
• Meta Ships Muse-Glimmer-30B: A 30B Dense Multimodal Model — NielsRogge · 2026-08-10
• Meta Releases Muse Glimmer: A 30B Open-Source Model Excellelling in Agentic Tasks — kimmonismus · 2026-08-10
• [source] unsloth Releases GGUF Quantized Version of Muse Glimmer 30B with Setup Guide — Nunki08 · 2026-08-10
• Meta Open-Sources Muse Glimmer: A 30B Vision-Language Model for Local Agents — ariG23498 · 2026-08-10
• Hugging Face Details Muse Glimmer: Built for Local Agents — ariG23498 · 2026-08-10
• Muse Glimmer Eval: 30B Model Beats Qwen and Gemma — kimmonismus · 2026-08-10
• Alexandr Wang Confirms Meta Will Release Open-Source Agentic Model Muse Glimmer — rohanpaul_ai · 2026-08-10
• Muse Glimmer 30B Weights Now Available on Hugging Face — ns123abc · 2026-08-10
• Meta Returns to Open Source: Muse Glimmer 30B Model and NVFP4 Quantized Version Released — Xianbao_QIAN · 2026-08-10
• Meta Releases Muse Glimmer: A 30B Open-Weight Local Agent Model — hsu_byron · 2026-08-10
• Muse Glimmer 30B Weights Released: Local, Agentic, Multimodal — tokenbender · 2026-08-10
• Meta Open-Sources 30B On-Device Agent Model, Balances Closed Frontier with Open Distillation — eyishazyer · 2026-08-10
• Meta Officially Releases Muse Glimmer: An Open Agentic Model That Runs on Your Device — aigclink · 2026-08-10
• vLLM Releases Deployment Guide for Muse-Glimmer-30B Local Multimodal Agent — vllm_project · 2026-08-10
• Meta Open-Sources 30B Agentic Model, Runs on a Single Consumer GPU — Justgototheeffinmoon · 2026-08-10
• 1M Token Context on Single RTX 3090 Achieved via KVarN Quantization — Anbeeld · 2026-08-10
• Meta Open-Sources MuseGlimmer: A 30B Local Agent Model — AGI Hunt · 2026-08-10
• Meta's New Open Multimodal Model Muse Glimmer Lands on Ollama — ollama · 2026-08-10
• Meta Open-Sources 30B Parameter Agent Model MuseGlimmer, Runs Locally on 24GB VRAM — AI寒武纪 · 2026-08-10
• Meta Launches Muse Glimmer: A 30B Open-Source Model for Local Agents — mchiang0610 · 2026-08-10
• Running Muse Glimmer Locally: Ollama Details 128K Context and Controllable Reasoning — mchiang0610 · 2026-08-10
• Weights for Meta's Open-Source Muse Glimmer Hit Hugging Face — jffwng · 2026-08-10
• Meta Releases Muse Glimmer: 30B Open-Weights Model for Local Agent Workflows — rohanpaul_ai · 2026-08-10
• Meta Open-Sources Muse-Glimmer-30B Multimodal Model Weights — jffwng · 2026-08-10
• 30B Model Hits 114 tps with Tuned Quants, Targeting <16G VRAM — tokenbender · 2026-08-10
• Meta Releases Open-Source 30B Agent Model MuseGlimmer, Runs on Mac and Consumer GPUs — APPSO · 2026-08-10
• Meta Drops Muse Glimmer 30B: A Multimodal Agentic Model for Local Deployment — mishig25 · 2026-08-10
• SGLang Announces Day-0 Support for Meta's Muse Glimmer, Hitting 230 tok/s on RTX 5090 — NVIDIAAI · 2026-08-10
• Running Meta's Muse Glimmer 30B Locally on Mac Studio at 30 tok/s — MaziyarPanahi · 2026-08-10
• Muse Glimmer 30B Gets Zero-Day Llama.cpp Support, Runs Locally on Mac Studio — MaziyarPanahi · 2026-08-10
• Meta Showcases Muse Glimmer: Model Autonomously Deploys to HF and Optimizes Inference — AIatMeta · 2026-08-10
• Meta Launches Muse Glimmer: 30B On-Device Agentic Model — PyTorch · 2026-08-10
• Running Muse Glimmer 30B with 256k Context on a Single RTX 3090: Benchmarks — coder543 · 2026-08-10
• 2-bit Quantized Muse Glimmer Calls 100+ Tools on 14GB RAM — danielhanchen · 2026-08-10
• Meta Releases Open-Source Agentic Model Muse Glimmer — TheZachMueller · 2026-08-10

5 near-duplicate retellings: danielhanchen · ns123abc · mervenoyann · pcuenq · sierracatalina