Aakib Ansari.
Get in touch
Back to articles
News Brief

Meta Releases Muse Glimmer: Apache 2.0 Licensed 30B Local Agent Model

Md Aakib Ansari
Md Aakib AnsariWeb Developer & AI Tools Reviewer
•2 min read•Model: Meta Muse Glimmer
Meta Releases Muse Glimmer: Apache 2.0 Licensed 30B Local Agent Model

Meta Superintelligence Labs released Muse Glimmer on August 10, 2026—a 30-billion-parameter open-weight model distilled from its flagship Muse Spark. Released under a permissive Apache 2.0 license, Muse Glimmer is engineered to handle complex, multi-step agentic workflows directly on local developer machines without requiring active cloud connections.

Unlike standard dense chat models, Glimmer is co-trained with its agent harness specifically for tasks requiring sustained tool use, terminal interactions, code execution, and file management. The model is multimodal (supporting text and image inputs) and features native failure recovery and state preservation. On consumer hardware, the 30B model is highly optimized; when quantized to 4-bit precision, it comfortably fits within 18–20 GB of VRAM, making it runnable on standard consumer GPUs and Apple Silicon Macs via integrations with llama.cpp, MLX, and ExecuTorch.

By open-sourcing a dedicated agent model, Meta is positioning itself to compete directly in the local developer space alongside specialized agents like KAT-Coder-V2.5. The release signals a strategic pivot back toward permissive open-source models, providing developers with a privacy-preserving alternative to closed cloud orchestration suites.

Frequently Asked Questions

What is Meta Muse Glimmer?
Muse Glimmer is a 30-billion parameter, open-weight agent model released by Meta. It is distilled from the larger Muse Spark model and licensed under Apache 2.0.
Can Muse Glimmer run locally?
Yes, it is designed specifically for offline, local execution. When quantized to 4-bit, it can run on consumer GPUs and Apple Silicon Macs with 18-20 GB VRAM.
What frameworks are supported?
Muse Glimmer natively supports developer frameworks like llama.cpp, MLX, and ExecuTorch for cross-platform deployment.

Related Articles

Ant Group's inclusionAI Team Releases Ling 3.0 Flash FP8 Under MIT License
News Brief2 min read
Ant Group's inclusionAI Team Releases Ling 3.0 Flash FP8 Under MIT License

Ant Group's inclusionAI team has released Ling 3.0 Flash FP8, a highly efficient 124-billion parameter Mixture-of-Experts (MoE) model. Featuring an MIT license and a custom hybrid attention architecture, the model reduces active parameters to 5.1 billion per token, matching the performance of much larger models while dramatically lowering operational costs.

Liquid AI's LFM2.5-2.6B Matches Models Three Times Its Size on Agentic Tasks
News Brief2 min read
Liquid AI's LFM2.5-2.6B Matches Models Three Times Its Size on Agentic Tasks

Liquid AI released LFM2.5-2.6B on August 4, a 2.69B-parameter on-device model purpose-built for agentic workloads. Using a hybrid architecture of short convolution blocks and grouped query attention, it runs under 2.5 GB of memory and reaches approximately 220 tokens/s on Apple M5 Max — while matching or exceeding Qwen3.5-9B on tool use and instruction-following benchmarks.

The White House Finalized Its AI Safety Vetting Framework. Open-Weight Models Are Exempt.
News Brief2 min read
The White House Finalized Its AI Safety Vetting Framework. Open-Weight Models Are Exempt.

The Trump administration finalized a voluntary pre-release cybersecurity vetting framework for frontier AI models on August 4, 2026, briefing OpenAI, Anthropic, Google, Meta, Nvidia, and Microsoft in a private session. The framework applies only to closed, proprietary models; open-weight models are explicitly exempt. NIST simultaneously released a draft of SP 800-239 on AI data center security.