llmrun
MetaMuse Spark

Muse Spark Open 816B A42B — Hardware Requirements & GPU Compatibility

Vision

Muse Spark Open is the open-weights release of Meta Superintelligence Labs' Muse Spark, a mixture-of-experts model with 816 billion total parameters and about 42 billion active per token. It is built for coding and agentic work, including multi-step reasoning, tool use, and long-horizon task completion, and it is natively multimodal: it takes text, images, video, and audio as input and produces text. The weights are large. Meta says it needs at least 700 GB of GPU memory for NVFP4 and about 2 TB for BF16. Community GGUF quants reach about 304 GB at Q2, which fits a 512 GB unified-memory machine, so this is server-class territory. The context length is 1 million tokens at inference. Reasoning depth is adjustable with low, medium, high, and xhigh effort levels, and Meta recommends high or xhigh for hard coding and agentic tasks. It is released under Meta's Muse Model License, a custom license, so read its terms before commercial use. This open version includes extra safety post-training.

822.0B params · MoE42B activeOther815 downloads443 likes3.5K quant downloads

Muse Spark Open 816B A42B runs on a multi-GPU server

Q4_K_M needs ~551.9 GB · about 8× H100 SXM 80 GB · or Mac Studio (M3 Ultra, 512GB) at IQ3_XXS (~25 tok/s)

Score 128.7#7 of 94 open#5 of 90 that fit ≤640 GBOther

Not set up for it?

Reference GPU · RTX 4090 24 GB

Estimates, ±20% · how we estimate

llmrun Composite Score →

128.7

Benchmarks · tier chart ↓
Among 94 open modelsHow the Composite Score is built

Run Muse Spark Open 816B A42B

One recommended way. Every other method, OS and quant is in the step-by-step guide.

Recommended

vLLM on a multi-GPU server
tensor + pipeline parallel · needs vLLM on every node
$vllm serve meta-models/Muse-Spark-Open-816B-A42B --max-model-len 16384 --tensor-parallel-size 8

Size the nodes from the table above

Download: Muse-Spark-Open-816B-A42B-Q4_K_M.gguf493.22 GBunsloth

install, other ways

Other ways

each opens step-by-step

Run in cloud

Rent 8× H100 SXM node · Q4_K_M

Estimated hourly price
~$17.30/hr
est., whole server
GPUs
8 GPUs
640 GB of VRAM
Deployon Vast.aiRent H100 SXM$2.16/hr per GPU (opens in new tab)
Compare GPUs →

How Much VRAM Does Muse Spark Open 816B A42B Need?

Select a quantization level; the listed hardware adheres to it.

QuantizationVRAM+32KRuns on
248.7 GB268.8 GB2-bit · i-quant extreme compression — Importance-weighted 2-bit, extreme compression — significant quality lossRuns on Instinct MI350X 288 GB
384.3 GB404.4 GB2-bit · K-quant — 2-bit quantization with K-quant improvementsRuns on 6× H100 SXM 80 GB
440.8 GB460.9 GB3-bit medium — 3-bit medium quantizationRuns on 7× H100 SXM 80 GB
452.1 GB472.2 GB4-bit · legacy — 4-bit legacy quantizationRuns on 7× H100 SXM 80 GB
542.5 GB562.7 GB4-bit medium — 4-bit medium quantization — most popular sweet spotRuns on 8× H100 SXM 80 GB
644.3 GB664.4 GB5-bit medium — 5-bit medium quantization — good quality/size tradeoffRuns on 6× H200 SXM 141 GB
746 GB766.1 GB6-bit, very good — 6-bit quantization, very good qualityRuns on 6× H200 SXM 141 GB
904.2 GB924.4 GB8-bit, near-lossless — 8-bit quantization, near-losslessRuns on 8× H200 SXM 141 GB
est.calculated, no published file yet

Selected Q4_K_M

4-bit medium quantization

551.9GBat 16K context

Weights
540.9 GB
KV cache
10.7 GB
Overhead
0.3 GB

Server-class: ~551.9 GB at 16K context · about 8× H100 SXM 80 GB

Which Hardware Can Run Muse Spark Open 816B A42B?

Q4_K_M · 542.5 GB

Muse Spark Open 816B A42B (Q4_K_M) requires 542.5 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 706+ GB is recommended. No consumer GPU has enough memory. 2 devices with unified memory can run Muse Spark Open 816B A42B, including NVIDIA DGX H100.

Needs server-class hardware

Muse Spark Open 816B A42B runs on a multi-GPU server · Q4_K_M needs ~551.9 GB

Hardware that runs this model
HardwareQuantMemorySpeedContextHardware page
Mac 512 GB · Mac Studio (M3 Ultra, 512GB)IQ3_XXS359.8 GB~25 tok/s32K ctxMac Studio →IQ3_XXS · 359.8 GB · ~25 tok/s · 32K ctx
8× H100 SXM 80 GB · 640 GBQ4_K_M551.9 GB—32K ctxH100 SXM →Q4_K_M · 551.9 GB · 32K ctx
5× H200 SXM 141 GB · 705 GBQ4_K_M551.9 GB—32K ctxH200 SXM →Q4_K_M · 551.9 GB · 32K ctx
4× B200 192 GB · 768 GBQ4_K_M551.9 GB—32K ctxB200 →Q4_K_M · 551.9 GB · 32K ctx

Also: NVIDIA DGX A100 640GB · NVIDIA DGX H100

≤ 640 GB tier →

By VRAM tier · typical GPUs and Macs at Q4_K_M

Consumer GPUs (up to 32 GB) and Macs up to 256 GB: too small for this model (needs ≥ 258.1 GB)8 GB tier →12 GB tier →16 GB tier →24 GB tier →Macs →

How good is Muse Spark Open 816B A42B?

The llmrun Score merges 5 public benchmark boards into a single figure, where 100 represents the average for open models.

Score vs VRAM in the ≤640 GB tier

  • Muse Spark Open 816B A42B
  • Other models
  • Best Score for its VRAM

#5 of 90 that fit ≤640 GB

Each dot is a model at the best quant that fits ≤640 GB. The line joins the models with the best Score for their VRAM; a dot below it is beaten by a model that needs no more memory.

Full ≤640 GB ranking →

Benchmarks

Here are all 6 boards that have results for this model, sorted by their weight in the llmrun Score; badges show who ran the test.

  • GPQA Diamond

    Epoch AI · Oct 2026Independent

    88.6%

    #12/98 open

    Independent
  • AIME 2024/2025

    Epoch AI · Oct 2026Independent

    98.1%

    #3/86 open

    Independent
  • FrontierMath

    Epoch AI · Oct 2026Independent

    69.8%

    #2/17 open

    Independent
  • LMArena WebDev

    LMArena · Oct 2026Arena

    1511.3

    #13/52 open

    Arena
  • SimpleQA

    Epoch AI · Oct 2026Independent

    56.6%

    #1/17 open

    Independent
  • FrontierMath Tier 4

    Epoch AI · Oct 2026Independent

    31.7%

    #2/13 open

    Independent
All 6 Muse Spark Open 816B A42B benchmarks

In the ≤640 GB class

90 models fit ≤640 GB

Same Score, only models that fit ≤640 GB at their best quant: this model's peers by memory need.

  1. #1

    MiMo V2.6 Pro RLXiaomi

    134.9

    614.1 GB

  2. #2

    MiMo V2.6 Flash RLXiaomi

    131.9

    186.5 GB

  3. #3

    GLM 5.3Z.ai

    130.2

    453.8 GB

  4. #4

    DeepSeek V4.1 FlashDeepSeek

    129.1

    458.4 GB

  5. #5

    Muse Spark Open 816B A42BMeta

    128.7

    551.9 GB

  6. #6

    DeepSeek V4 Flash 0731DeepSeek

    127.6

    183.0 GB

Full ≤640 GB ranking →

Alternatives to Muse Spark Open 816B A42B

Official models with an llmrun Score, each answering one question; sizes are each model's default quant (usually Q4_K_M) with a 16K-token context.

Similar models

10
ModelScoreVRAMWhy it is similar
MiMo V2.6 Flash RLXiaomi131.9186.5 GBsimilar score
DeepSeek V4.1 FlashDeepSeek129.1458.4 GBsimilar score
DeepSeek V4 Flash 0731DeepSeek127.6183 GBsimilar score
GLM 5.3 FlashZ.ai124.1205.2 GBsimilar score
GLM 5.2Z.ai122.9453.8 GBsimilar score
Qwen3.8 27BAlibaba121.517.9 GBsimilar score
Kimi K2.6Moonshot AI121.1617.6 GBsimilar score
GLM 5.1Z.ai117.7454.1 GBsimilar score
DeepSeek V4 ProDeepSeek116.5959.8 GBsimilar score
MiMo V2.5 ProXiaomi116.5616 GBsimilar score

Specifications

Publisher
Meta
Family
Muse Spark
Parameters
822.0B
Active parameters
42B per token
Release Date
2026-09-15
License
Other

Resources

Frequently Asked Questions

How much VRAM does Muse Spark Open 816B A42B need?

Muse Spark Open 816B A42B requires 542.5 GB of VRAM at Q4_K_M, or 1808.5 GB at BF16.

VRAM = Weights + KV Cache + Overhead

Weights = 822.0B × 4.8 bits ÷ 8 = 493.2 GB

KV Cache + Overhead ≈ 49.3 GB (at 2K context + ~0.3 GB framework)

Fit ratings and hardware model lists check this model with room for a 16K-token context, which needs a little more memory.

VRAM usage by quantization

542.5 GB

Learn more about VRAM estimation →

Can NVIDIA GeForce RTX 5090 run Muse Spark Open 816B A42B?

No — Muse Spark Open 816B A42B requires at least 248.7 GB at IQ2_XXS, which exceeds the NVIDIA GeForce RTX 5090's 32 GB of VRAM.

What's the best quantization for Muse Spark Open 816B A42B?

For Muse Spark Open 816B A42B, Q4_K_M (542.5 GB) offers the best balance of quality and VRAM usage. Q5_K_M (644.3 GB) provides better quality if you have the VRAM. The smallest option is IQ2_XXS at 248.7 GB.

VRAM requirement by quantization

IQ2_XXS
248.7 GB
Q2_K
384.3 GB
Q4_K_S
508.6 GB
Q4_K_M ★
542.5 GB
Q5_K_M
644.3 GB
BF16
1808.5 GB

★ Recommended — best balance of quality and VRAM usage.

Learn more about quantization →

Can I run Muse Spark Open 816B A42B on a Mac?

Yes, but only at lower quantizations. The smallest Mac that can run Muse Spark Open 816B A42B is Mac Studio (M3 Ultra, 512GB) at IQ2_XXS; 1 of the 156 Macs we list can run it at some quantization. For Q4_K_M (542.5 GB) you need a Mac with more unified memory.

Can I run Muse Spark Open 816B A42B locally?

Yes — Muse Spark Open 816B A42B can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.

What's the download size of Muse Spark Open 816B A42B?

At Q4_K_M, the download is about 493.22 GB. The full-precision BF16 version is 1644.07 GB. The smallest option (IQ2_XXS) is 226.06 GB.

Which GPUs can run Muse Spark Open 816B A42B?

No single consumer GPU has enough VRAM to run Muse Spark Open 816B A42B at Q4_K_M (542.5 GB). Multi-GPU or professional hardware is required.

Which devices can run Muse Spark Open 816B A42B?

2 devices with unified memory can run Muse Spark Open 816B A42B at Q4_K_M (542.5 GB), including NVIDIA DGX A100 640GB, NVIDIA DGX H100. Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.