How much VRAM does Muse Spark Open 816B A42B need?
Muse Spark Open 816B A42B needs 464 GB for its weights at 4-bit, 466 GB with an 8K context and 468 GB with 32K. At 8-bit, 815 GB.
- Parameters
- 822B
- Active per token
- 42B
- Longest context
- 1M
- Released
- Sep 2026
- License
- Muse Model License
- Strongest at
- Chat, Docs
Memory at each precision
| Precision | Weights | With 8K | With 32K | With 128K |
|---|---|---|---|---|
| 16-bit | 1,531 GB | 1,533 GB | 1,535 GB | 1,541 GB |
| 8-bit | 813 GB | 815 GB | 817 GB | 823 GB |
| 6-bit | 628 GB | 630 GB | 631 GB | 637 GB |
| 5-bit | 545 GB | 546 GB | 548 GB | 554 GB |
| 4-bit | 464 GB | 466 GB | 468 GB | 474 GB |
| 3-bit | 374 GB | 376 GB | 378 GB | 384 GB |
| 2-bit | 287 GB | 289 GB | 291 GB | 297 GB |
| 1.6-bit | 158 GB | 160 GB | 161 GB | 167 GB |
One conversation, 16-bit context cache, runtime overhead included.
What runs it
| Machine | Best fit | Context | Speed | First reply |
|---|---|---|---|---|
| RTX 3060 12 GB | Too big | |||
| RTX 4060 Ti 16 GB | Too big | |||
| RTX 3090 | Too big | |||
| RTX 4090 | Too big | |||
| RTX 5090 | Too big | |||
| 2× RTX 3090 | Too big | |||
| RTX PRO 6000 Blackwell | Too big | |||
| M4 Max 128 GB | Too big | |||
| M5 Max 128 GB | Too big | |||
| Ryzen AI Max+ 395 128 GB | Too big | |||
| DGX Spark 128 GB | Too big | |||
| M3 Ultra 256 GB | 1.6-bit | 352K | 40 tok/s | 26 s |
| M5 Ultra 512 GB | 2-bit | 1M | 35 tok/s | 8.1 s |
| H100 SXM | Too big | |||
| DGX Station GB300 | 1.6-bit | 1M | 183 tok/s | 0.7 s |
| 8× H200 NVL | 8-bit | 1M | 59 tok/s | 0.4 s |
Best fit is the highest quality, up to 8-bit, that leaves 16K of context. First reply is the wait for a 2K prompt. Speeds are rough, plus or minus 25%.
How to run it
- llama.cpp
llama-server -hf unsloth/Muse-Spark-Open-816B-A42B-GGUF- vLLM
vllm serve meta-models/Muse-Spark-Open-816B-A42B- MLX (Mac)
mlx_lm.chat --model mlx-community/Muse-Spark-Open-816B-A42B-4bit
Questions
How much VRAM does Muse Spark Open 816B A42B need?
Muse Spark Open 816B A42B needs 464 GB for its weights at 4-bit, 466 GB with an 8K context and 468 GB with 32K. At 8-bit, 815 GB.
Can Muse Spark Open 816B A42B run on an RTX 4090?
Not on one card. The smallest machine here that runs it is the M3 Ultra 256 GB.