Muse Glimmer 30B has an official 24GB local-AI path, but Meta’s dynamic Q4 build targets 32GB hardware. The smaller Q4_K_M text model is 16.8GB. The dynamic Q4_K_XL text model is 19.7GB, and optional multimodal and speculative-decoding components add separate files.
The official memory ladder
| Path | Published file or rough memory | Official hardware target |
|---|---|---|
| Q4_K_M text only | 16.8GB file; roughly 17GB memory | 24GB |
| Q4_K_M with vision | Roughly 19GB including the 1.4GB vision encoder | 24GB |
| Q4_K_M with vision and DFlash | Roughly 20GB including the 1.6GB drafter | 24GB |
| Dynamic Q4_K_XL text only | 19.7GB file; roughly 20GB memory | 32GB |
| Dynamic Q4_K_XL with vision and DFlash | Roughly 23GB | 32GB |
| BF16 | Full-precision deployment path | 64GB |
Why 16GB is not the official target
A 16.8GB model file is already larger than a nominal 16GB device before runtime allocations, KV cache, vision processing or the optional drafter. CPU/GPU offload may make a mixed-memory configuration possible, but that is different from a full-GPU fit and is not Meta’s published 24GB target.
Context is a ceiling, not a reservation
The model’s native context ceiling is 131,072 tokens. That does not mean every local run should allocate the maximum. Practical memory use depends on the runtime, context, cache representation, image inputs, concurrency and whether speculative decoding is enabled. Meta also identifies llama.cpp b10353 as the minimum recognized build, so an older binary can fail for reasons unrelated to capacity.
Use the full planning guide
The complete Muse Glimmer 30B VRAM guide separates file size, rough working memory and hardware targets, then provides a validation checklist for exact artifacts and runtime settings.
- Compare current 24GB hardware paths
- Compare current 32GB hardware paths
- Choose between 24GB and 32GB hardware for Muse Glimmer \n
- Run the official GGUF path with llama.cpp
- Stage a reproducible Muse Glimmer benchmark record
- Review current screened 24GB and 32GB offers — saved observations, not guaranteed stock. \n
- Stage a private 24GB-or-32GB hardware watchlist
- Subscribe to the main RSS feed
Primary sources
- Meta Muse Glimmer 30B GGUF repository
- Meta Muse Glimmer 30B model repository
- Meta Muse Glimmer deployment documentation
- Meta Muse Glimmer model overview
Checked October 25, 2026. File size and rough working-memory figures are not speed measurements. Hardware capacity alone does not prove runtime support, usable context, multimodal throughput or stability.