Muse Glimmer 30B
Muse Glimmer 30B needs 18.7 GB of VRAM at Q4_K_M with 8k context, computed from its config.json as checked on 16 October 2026. The smallest card that holds it has 24 GB, and the fastest desktop card that runs it, the RTX 5090, generates about 73 tokens per second.
- Parameters
- 29.8B
- Context
- 131,072 tokens
- Licence
- Apache 2.0
- Released
- August 2026
- Config checked
- 16 October 2026
Open Muse Glimmer 30B in the LLM checker
Which card size runs it
18.7 GB against each memory size. A model may use 95% of a card, and only what is left over, or missing, is coloured; tight means above 85%, with little room for a longer conversation.
Other quantisations, context lengths or a Q8 KV cache: the VRAM calculator.
Computed 16 October 2026Which GPU to pick
The smallest card here that loads it at Q4_K_M with 8k of context is the RX 7900 XT (20 GB): 18.7 GB needed, an estimated 32 tokens per second, and a tight fit that leaves little for the conversation. For room to work, meaning 32,768 tokens of context while staying under 85% of memory, step up to the RTX 3090 (24 GB), which needs 19 GB for that and generates at about 38 tokens per second.
At Q8_0, which is near-lossless, the card is the RTX A6000 (48 GB), needing 33.5 GB. On Apple silicon, a Mac mini M4 Pro holds it with 32,768 tokens of context at an estimated 11 tokens per second; prompt processing is slower than on Nvidia, which long documents make obvious. Among unified-memory desktops, the Ryzen AI Max+ 395 holds it at 32,768 tokens in 19 GB, at an estimated 10 tokens per second.
Speed on each card
| GPU | Fits | Est. tokens/s | Feels like | Memory | Bandwidth |
|---|---|---|---|---|---|
| H100 SXM | Yes | ≤ 136 | faster than you can read | 80 GB | 3,350 GB/s |
| A100 | Yes | 83 | faster than you can read | 80 GB | 2,039 GB/s |
| RTX 5090 | Yes | 73 | faster than you can read | 32 GB | 1,792 GB/s |
| RTX PRO 6000 Blackwell | Yes | 73 | faster than you can read | 96 GB | 1,792 GB/s |
| RTX PRO 5000 Blackwell | Yes | 54 | comfortable for chat | 48 GB | 1,344 GB/s |
| RTX 4090 | Yes | 41 | comfortable for chat | 24 GB | 1,008 GB/s |
| RX 7900 XTX | Yes | 39 | comfortable for chat | 24 GB | 960 GB/s |
| RTX 6000 Ada | Yes | 39 | comfortable for chat | 48 GB | 960 GB/s |
| RTX 3090 | Yes | 38 | comfortable for chat | 24 GB | 936 GB/s |
| RTX PRO 4500 Blackwell | Yes | 36 | comfortable for chat | 32 GB | 896 GB/s |
| L40S | Yes | 35 | comfortable for chat | 48 GB | 864 GB/s |
| RX 7900 XT | Tight | 32 | comfortable for chat | 20 GB | 800 GB/s |
| RTX A6000 | Yes | 31 | comfortable for chat | 48 GB | 768 GB/s |
| RTX PRO 4000 Blackwell | Yes | 27 | comfortable for chat | 24 GB | 672 GB/s |
| Radeon AI PRO R9700 | Yes | 26 | comfortable for chat | 32 GB | 640 GB/s |
| Arc Pro B70 | Yes | 25 | usable, a little slow | 32 GB | 608 GB/s |
| Arc Pro B60 | Yes | 18 | usable, a little slow | 24 GB | 456 GB/s |
| RTX PRO 4000 Blackwell SFF | Yes | 17 | usable, a little slow | 24 GB | 432 GB/s |
| RTX Spark | Yes | 12 | usable, a little slow | 128 GB (102 for the GPU) | 300 GB/s |
| DGX Spark | Yes | 11 | slow; fine for batch jobs | 128 GB (126 for the GPU) | 273 GB/s |
| Ryzen AI Max+ 395 | Yes | 10 | slow; fine for batch jobs | 128 GB (96 for the GPU) | 256 GB/s |
| RTX 4060 Ti | No | - | does not fit | 16 GB | 288 GB/s |
| RTX 5060 Ti | No | - | does not fit | 16 GB | 448 GB/s |
| RTX 5080 | No | - | does not fit | 16 GB | 960 GB/s |
| RTX 5070 Ti | No | - | does not fit | 16 GB | 896 GB/s |
| RTX 4080 SUPER | No | - | does not fit | 16 GB | 736 GB/s |
| RX 9070 XT | No | - | does not fit | 16 GB | 640 GB/s |
| RX 9060 XT | No | - | does not fit | 16 GB | 320 GB/s |
| RTX 3060 | No | - | does not fit | 12 GB | 360 GB/s |
| RTX 4070 | No | - | does not fit | 12 GB | 504 GB/s |
| RTX 5070 | No | - | does not fit | 12 GB | 672 GB/s |
| Arc B580 | No | - | does not fit | 12 GB | 456 GB/s |
Single-stream decode, from memory bandwidth at 70% efficiency. Prompt processing and batching not included. These are ceilings, not measurements: Muse Glimmer 30B reads only 17.3 GB of weights per token at Q4, so little that the per-token costs the formula leaves out decide much of the real speed. The figure marked ≤ is where a real runtime falls furthest below the number shown. See the speed estimator for other quantisations and Apple hardware, or every card compared if you are choosing hardware rather than a model.
Computed 16 October 2026Memory by quantisation and context
| Quant | 4,096 ctx | 8,192 ctx | 32,768 ctx |
|---|---|---|---|
| Q8_0 | 33.5 GB · fits 48 GB | 33.5 GB · fits 48 GB | 33.9 GB · fits 48 GB |
| Q5_K_M | 22.6 GB · fits 24 GB | 22.7 GB · fits 24 GB | 23 GB · fits 32 GB |
| Q4_K_M | 18.6 GB · fits 24 GB | 18.7 GB · fits 24 GB | 19 GB · fits 24 GB |
Weights at this quant: 17.3 GB at Q4. Every extra 1,000 tokens of context adds about 0.013 GB of KV cache at FP16, on top of a fixed 0.082 GB that the sliding-window layers hold once their window is full. Try other settings in the VRAM calculator.
About the model
Muse Glimmer 30B has 29.8B parameters. It has 52 layers: 39 use sliding-window attention over the last 2,048 tokens and 13 are full attention, all with 2 KV heads of dimension 128. Meta's 30B model from August 2026, distilled from Muse Spark and built for agents that run on consumer hardware. A dense model with a perception encoder; three of every four layers use a 2,048-token sliding window, so a long conversation costs very little memory.
Text and images 131,072-token window Apache 2.0 August 2026
The parameter count includes the vision encoder; a text-only GGUF is slightly smaller.
Native context window: 131,072 tokens. Licence: Apache 2.0, a permissive licence. Weights published in August 2026.
Good at
- Agents on a 24 GB card: Meta built it for local tool use, and its sliding-window cache keeps a 64k context cheap.
- Tool calling that recovers from failed calls, which the model card names as a design goal.
- It ships with a small drafter model for speculative decoding, which Meta says makes generation faster.
- Apache 2.0.
Watch out for
- The speed figures here are for the main model alone; the drafter adds its own memory, and its speed-up is not modelled.
- A runtime without a sliding-window cache stores every layer at full length and needs far more than these figures at long context.
- The native window is 131,072 tokens, half of what the newest Qwen and Gemma models offer.
- It is a new architecture: check that your runtime supports it before downloading.
Similar models
The models that need about the same memory as Muse Glimmer 30B at Q4_K_M and 8k, which makes them the real alternatives on whatever card you have. The context column is what each extra 1,000 tokens costs, and it is where models of the same size differ most.
| Model | Size | Needs | Per 1k context | Window | Released |
|---|---|---|---|---|---|
| Muse Glimmer 30B | 29.8B | 18.7 GB | 0.013 GB | 128k | 2026-08 |
| Gemma 3 27B | 27.4B | 18.1 GB | 0.082 GB | 128k | 2025-03 |
| Qwen3.8 27B | 27.8B | 18 GB | 0.066 GB | 256k | 2026-08 |
| Qwen3.6 27B | 27.8B | 18 GB | 0.066 GB | 256k | 2026-04 |
| Nemotron 3.5 Lightning 30B-A3B (MoE) | 31.6B · 3B active | 19.7 GB | 0.006 GB | 256k | 2026-08 |
See every model compared, which models fit on 8 to 96 GB, or the cheapest hour of every cloud card.
Questions
- Can I run Muse Glimmer 30B on a 24 GB card?
- Yes, at Q4_K_M and 8k context it needs about 18.7 GB, which fits a 24 GB card with room.
- How much VRAM does Muse Glimmer 30B need at Q8?
- About 33.5 GB at 8k context, or 33.9 GB at 32k. Q8 is near-lossless; use it when it fits.
- What is the context length of Muse Glimmer 30B?
- 131,072 tokens natively. Holding all of it at Q4_K_M takes about 20.3 GB, of which 1.8 GB is context.
- Why does Muse Glimmer 30B need so little memory for long context?
- Because most of its layers do not keep a cache that grows. It has 52 layers: 39 use sliding-window attention over the last 2,048 tokens and 13 are full attention, all with 2 KV heads of dimension 128. At 32k tokens the context costs about 0.5 GB; if all 52 layers kept a conventional full-length cache, the same conversation would cost about 1.7 GB. That saving depends on the runtime implementing the sliding-window cache. llama.cpp and vLLM do; a build that does not will use far more.
- Can Muse Glimmer 30B run on a 24 GB card?
- Yes, at Q4_K_M, with room for a long context: three of every four layers only cache the last 2,048 tokens, so the conversation adds little memory. The tables above show the exact figures for each card.
- How fast is Muse Glimmer 30B on an RTX 4090?
- At most about 41 tokens per second at Q4, single stream, which is comfortable for chat. That is a ceiling worked out from the card's memory bandwidth, not a measurement.
- Architecture values from config.json in meta-models/Muse-Glimmer-30B on Hugging Face, checked 16 October 2026. Every figure on this page is computed from those values; check them before buying hardware for this model.
- Estimates assume a single conversation on a card that is otherwise free. A desktop on the same GPU takes 0.5 to 2 GB.