Skip to main content
Models & Technology

Meta Open-Sources 30B AI Model Muse Glimmer, 4-bit Quantization Runs on a Single GPU with 24GB VRAM

Meta today released the Muse Glimmer AI model and open-sourced its model weights under the Apache 2.0 License.

IT HomeAbout 3 min read
Meta Open-Sources 30B AI Model Muse Glimmer, 4-bit Quantization Runs on a Single GPU with 24GB VRAM

Meta today released the Muse Glimmer AI model and open-sourced its model weights under the Apache 2.0 License. Muse Glimmer is a 30-billion-parameter (30B) model deeply optimized for local agent workflows designed to run continuously around the clock.

According to Meta, Muse Glimmer uses 4-bit quantization, allowing it to run locally on Macs or PCs with 24GB/32GB of VRAM. Tests confirmed that this compression has a negligible impact on agent task performance, with virtually no degradation.

Meta tested the model on MacBook M4 Max, M5 Max, and RTX 5090 platforms. The results show that its speed is sufficient for smooth conversations and real-time agent interactions, with all computation completed 100% locally on the device.

Meta Open-Sources 30B AI Model Muse Glimmer, 4-bit Quantization Runs on a Single GPU with 24GB VRAM

Compared with Gemma4-31B and Qwen3.6-27B, Muse Glimmer demonstrates strong performance for a model of its size across multiple mainstream large language model benchmarks.

Meta Open-Sources 30B AI Model Muse Glimmer, 4-bit Quantization Runs on a Single GPU with 24GB VRAM

Open-source link:

https://huggingface.co/meta-models/Muse-Glimmer-30B