Muse Spark Open (816B-A42B)
Meta's 816B MoE (42B active), 1M context, natively multimodal, gated Muse Model License. Meta quotes 700 GB VRAM for NVFP4; community runs on dual 512 GB Mac Studios.
Recommended quants
- Unsloth UD-IQ1_S (175.6 GB) / UD-IQ2_XXS (225.6 GB)
unsloth/Muse-Spark-Open-816B-A42B-GGUFExtreme low-bit; quality unmeasured. - UD-Q2_K_XL (311 GB) / UD-Q4_K_XL (505.5 GB)
unsloth/Muse-Spark-Open-816B-A42B-GGUF - MLX 3-bit
mlx-community/Muse-Spark-Open-816B-A42B-3bit - Official BF16 / NVFP4
meta-models/Muse-Spark-Open-816B-A42B(-NVFP4)Gated: accept license on HF.
Chat template Unverified
Reasoning effort high/xhigh recommended for complex work.
Fix: Check the official repo template after accepting the license.
Sampling settings
| Official | temp 1.0, top_p 1.0, top_k disabled; reasoning effort high or xhigh for coding/agents |
|---|
Runtimes
| vLLM | Unverified | Reported supported in launch coverage. |
|---|---|---|
| llama.cpp | Works | Unsloth GGUFs. |
| Ollama | Unverified | Library page muse-spark-open exists. |
| LM Studio | Unverified | |
| MLX | Works | mlx-community 3-bit. |
Tool calling in agent tools
| Claude Code (local endpoint) | Unverified | |
|---|---|---|
| OpenCode | Unverified | |
| Cline | Unverified | |
| Aider | Unverified |
Known issues
Sources checked
- https://huggingface.co/unsloth/Muse-Spark-Open-816B-A42B-GGUF
- https://huggingface.co/meta-models/Muse-Spark-Open-816B-A42B
Spotted something wrong or a new fix? Corrections are reviewed before publishing. (Submission form coming soon.)