DownloadMuse Glimmer-30B is Meta Superintelligence Lab’s open-weight multimodal model built specifically for autonomous agentic tasks on consumer hardware. Distilled from the larger Muse Spark, it combines multi-step reasoning, reliable tool use, coding, failure recovery, and image understanding in a dense 29.6B-parameter architecture with a dedicated 1.8B-parameter perception encoder. It supports more than 100 languages and a 131K+ token context window, allowing agents to maintain coherent plans across extended workflows. Muse Glimmer can interpret screenshots, charts, documents, and images alongside text, while configurable reasoning strength lets developers balance response quality and speed. Its quantized variants reduce the model below 20 GB for operation on systems with 24–32 GB of memory, and DFlash speculative decoding can substantially accelerate generation.
Features
- 29.6B-parameter dense causal Transformer architecture
- Text and image input with dedicated 1.8B perception encoder
- 131K+ token context window for long agent workflows
- Reliable tool calling and multi-step autonomous task execution
- Failure detection and recovery during complex workflows
- Four configurable reasoning levels from low to xhigh
- Quantized variants designed for 24–32 GB consumer hardware
- DFlash speculative decoding for significantly faster generation
Project Samples
