Muse Spark Open runs at 18 tok/s on a 512 GB M3 Ultra, a published figure reported by huggingface.co (llama.cpp, Q2_K). It needs about 342 GB of unified memory. It is Meta's 816B mixture-of-experts model (42B active), released under the Muse Model License licence.
Want to run Muse Spark Open faster — or step up to bigger models? Find your Mac →This is a published figure: a real run, reported by its source and linked below. The index records it as given and never adjusts it.
| Metric | Value |
|---|---|
| Tokens per second | sourced18 tok/s |
| Provenance | published figure · huggingface.co |
| Quantization | Q2_K |
| Engine | llama.cpp |
| Figure dated | 2026-10-15 |
| Minimum RAM | 342 GB |
| Parameters | 816B MoE (42B active) |
| Chip bandwidth | 819 GB/s |
Each row as the index holds it, with its source. “Mac RAM” is the memory of the Mac a run used, not the model’s requirement (342 GB). The headline figure follows the reference rule: the lower-median measured plain-decode run at 5 bits or fewer (a run whose quant was not stated counts only when no stated-quant run qualifies), else the estimate; a run marked “not used to anchor estimates” is listed here but never becomes the headline.
| Speed | Quant | Engine | Mac RAM | Context | Source | Note |
|---|---|---|---|---|---|---|
| sourced21 tok/s | oQ3e | MLX | 512 GB | 1024-token prompt, 128 generated | huggingface.co | Hugging Face launch post: MLX (oMLX/mlx-lm, app not named); the pre-16-Oct oQ3e upload (369.5 GB = 344 GiB; the 16 Oct re-upload is 371.5 GB, 3.62 bpw), ~225 tok/s prefill. |
| sourced18 tok/s | Q2_K | llama.cpp | 512 GB | — | huggingface.co | Hugging Face launch post: llama serve -hf ggml-org/Muse-Spark-Open-816B-A42B-GGUF:Q2_K (imatrix, 304 GB); prompt length not stated. |
Ollama — the index has not verified an Ollama tag for Muse Spark Open, so it doesn’t print one (a guessed tag fails with “pull model manifest: file does not exist”). Search the Ollama library →
MLX — pick a build from Hugging Face’s MLX listings, then run it with its repo id:
Downloads: Hugging Face repo → · model card →
The 18 tok/s figure above came from llama.cpp at Q2_K; other runtimes and quants run at different speeds.
Prefer a GUI? LM Studio downloads GGUF and MLX builds in one click. Step by step: install Ollama on a Mac.
| Chip | Speed | Quant | Engine | Bandwidth |
|---|---|---|---|---|
| M5 Ultra | estimated31 tok/s | oQ3e | MLX | 1228 GB/s |
Meta's open-weight Muse Spark (weights 15 Oct 2026): an 816B MoE with 42B active, 1M context and text, image, video and audio input, under the Muse Model License (products must show a 'Built with Muse Spark Open' credit; API hosts above $20M revenue need a separate agreement). One 512 GB Mac Studio runs it at 2–3-bit: Hugging Face's launch post reports 21 tok/s (oQ3e, MLX) and 18 tok/s (Q2_K, llama.cpp b11671+) on an M3 Ultra, while 4-bit needs two Macs. Ollama and LM Studio offer it cloud-only.
Muse Model License licence · 816B MoE · 1000K context · LLMCheck capability 49/50 · see its leaderboard rank →
Muse Spark Open needs about 342 GB of unified memory. These current Apple Silicon Macs have the headroom to run it comfortably:
As an Amazon Associate, LLMCheck earns from qualifying purchases. These affiliate links cost you nothing extra and help keep the LLMCheck index free.
See how Muse Spark Open stacks up against other models on your specific Mac hardware.
Open Compare Tool Full Leaderboard