Open-weight model · Text generation
Muse-Spark-Open-816B-A42B-MXFP4
by Intel AI intel-ai/Muse-Spark-Open-816B-A42B-MXFP4
Muse-Spark-Open-816B-A42B-MXFP4 is an open-weight model for text generation from Intel AI, released under other. It has 822B parameters and a 1,048,576-token context. At 16-bit it needs about 1972.9 GB of GPU memory, which fits on 8x MI325X from $16.00 an hour; at 4-bit, 493.2 GB on 2x MI325X from $4.00, at the lowest prices in the SAVRN Index. It draws 50 downloads a month.
This model is an MXFP4 (W4A4, groupsize 32) model of meta-models/Muse-Spark-Open-816B-A42B generated by intel/auto-round using RTN. Please follow the license of the original model.
Parameters822B
Context1,048,576
Weights508.9 GB
Licenseother
AccessOpen weights
Monthly Downloads50
Runs On
What it takes to serve Muse-Spark-Open-816B-A42B-MXFP4 (822B parameters): the memory its weights need at each precision, and the cheapest way to rent enough data-center GPUs to hold them.
| Precision | Weights | Memory needed | Cheapest setup | Per hour | Also fits |
|---|---|---|---|---|---|
| 16-bit | 1644.1 GB | 1972.9 GB | 8x MI325X (256 GB) Vultr |
$16.00 | 7x MI355X $18.13 · 8x B300 $52.80 |
| 8-bit | 822.0 GB | 986.4 GB | 4x MI325X (256 GB) Vultr |
$8.00 | 4x MI355X $10.36 · 6x MI300X $11.10 |
| 4-bit | 411.0 GB | 493.2 GB | 2x MI325X (256 GB) Vultr |
$4.00 | 2x MI355X $5.18 · 3x MI300X $5.55 |
Memory is the weights at that precision plus 20% for the runtime and a short context; a long context needs more. Prices are the lowest on-demand hourly rates in the SAVRN Index, read Oct 27, 2026.
Muse-Spark-Open-816B-A42B-MXFP4 on every accelerator the SAVRN Index prices, at every precision
Model Card
This model is an MXFP4 (W4A4, groupsize 32) model of meta-models/Muse-Spark-Open-816B-A42B generated by intel/auto-round using RTN. Please follow the license of the original model. The model can produce factually incorrect output, and should not be relied on to produce factually accurate information. Because of the limitations of the pretrained model and the finetuning datasets, it is possible that this model could generate lewd, biased or otherwise offensive outputs. Therefore, before deploying any applications of the model, developers should perform safety testing. Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. Here are a couple…
Excerpt from the card by Intel AI, licensed other.
Configuration
- Architecture
- MuseSparkForConditionalGeneration
- Context length (tokens)
- 1,048,576
- Layers
- 62
- Hidden size
- 8,192
- Feed-forward size
- 16,384
- Attention heads
- 128
- Key/value heads
- 16
- Head dimension
- 64
- Vocabulary size
- 202,048
- Experts
- 256
- Experts active per token
- 8
- Sliding window (tokens)
- 2,048
- Model type
- muse_spark
- Quantization
- auto-round
Identity and Version
- Repository
- intel-ai/Muse-Spark-Open-816B-A42B-MXFP4
- Publisher
- Intel AI
- Task
- Text generation
- Modality
- Text
- Library
- transformers
- Parameters
- 822B parameters
- Languages
- Not stated by the source
- Revision
- ba18db16ab95e0870796d315205c7e6e587f5fbe
- First published
- 2026-10-14
- Last updated
- 2026-10-27
Files and Weights
79 files, 508.9 GB in total. The weights are 65 files totalling 508.9 GB in safetensors.
Weights65 files · 508.9 GB
Configuration6 files · 11.5 MB
Tokenizer2 files · 28.5 MB
Documentation3 files · 16.2 KB
Other2 files · 3.6 MB
Repository1 file · 1.6 KB
Every file
| File | Type | Size | SHA-256 |
|---|---|---|---|
| model-00001-of-00065.safetensors | Weights | 10.7 GB | f90ffb8bcb73 |
| model-00002-of-00065.safetensors | Weights | 550.3 MB | 340af63b0b7f |
| model-00003-of-00065.safetensors | Weights | 7.4 GB | 9ca78abced9f |
| model-00004-of-00065.safetensors | Weights | 7.4 GB | a8aa19a24561 |
| model-00005-of-00065.safetensors | Weights | 7.4 GB | 907133a550fa |
| model-00006-of-00065.safetensors | Weights | 7.4 GB | 8660db81d38e |
| model-00007-of-00065.safetensors | Weights | 10.7 GB | 6ec7959539e0 |
| model-00008-of-00065.safetensors | Weights | 7.4 GB | 673595250285 |
| model-00009-of-00065.safetensors | Weights | 26.4 GB | b22381b8d5a7 |
| model-00010-of-00065.safetensors | Weights | 7.4 GB | e47d333fa361 |
| model-00011-of-00065.safetensors | Weights | 7.4 GB | ab18c2fdd6dc |
| model-00012-of-00065.safetensors | Weights | 7.4 GB | 3cc8b281a856 |
| model-00013-of-00065.safetensors | Weights | 7.4 GB | 6e67ce15fd9b |
| model-00014-of-00065.safetensors | Weights | 7.4 GB | a3cbcd2cb90e |
| model-00015-of-00065.safetensors | Weights | 7.4 GB | 9825ab24c04c |
| model-00016-of-00065.safetensors | Weights | 7.5 GB | 515d3cfc9256 |
| model-00017-of-00065.safetensors | Weights | 7.4 GB | 9bf6f520d55d |
| model-00018-of-00065.safetensors | Weights | 7.4 GB | 83bd2bc0e737 |
| model-00019-of-00065.safetensors | Weights | 7.4 GB | 67bd6bd266e9 |
| model-00020-of-00065.safetensors | Weights | 7.4 GB | 0a9ca25f7407 |
| model-00021-of-00065.safetensors | Weights | 7.4 GB | 9bdf5df02bba |
| model-00022-of-00065.safetensors | Weights | 7.5 GB | 0d881423b321 |
| model-00023-of-00065.safetensors | Weights | 7.4 GB | f0885e8cc39c |
| model-00024-of-00065.safetensors | Weights | 7.5 GB | 1dae8c13a58b |
| model-00025-of-00065.safetensors | Weights | 7.4 GB | afd9f1adeb8a |
| model-00026-of-00065.safetensors | Weights | 7.5 GB | 62bc346a7152 |
| model-00027-of-00065.safetensors | Weights | 7.4 GB | e98c8e364615 |
| model-00028-of-00065.safetensors | Weights | 7.4 GB | b49bc5c02403 |
| model-00029-of-00065.safetensors | Weights | 7.4 GB | 0a38ee578a44 |
| model-00030-of-00065.safetensors | Weights | 7.4 GB | df03628b4691 |
| model-00031-of-00065.safetensors | Weights | 7.5 GB | 8d951595a1eb |
| model-00032-of-00065.safetensors | Weights | 7.4 GB | b9cfaf5f98b2 |
| model-00033-of-00065.safetensors | Weights | 7.4 GB | 9f7e14f93f53 |
| model-00034-of-00065.safetensors | Weights | 7.4 GB | 957936ef8c70 |
| model-00035-of-00065.safetensors | Weights | 7.5 GB | 05b8c015b684 |
| model-00036-of-00065.safetensors | Weights | 7.5 GB | ea7d9d7bbabb |
| model-00037-of-00065.safetensors | Weights | 7.4 GB | 7349adc3505b |
| model-00038-of-00065.safetensors | Weights | 7.4 GB | 4ebbaeec4ff3 |
| model-00039-of-00065.safetensors | Weights | 7.4 GB | 9af821328d50 |
| model-00040-of-00065.safetensors | Weights | 7.4 GB | 1fea09e61906 |
| model-00041-of-00065.safetensors | Weights | 7.5 GB | 9a98a0a94bcb |
| model-00042-of-00065.safetensors | Weights | 7.5 GB | 5ad29810542c |
| model-00043-of-00065.safetensors | Weights | 7.4 GB | 2a51fdd3cde9 |
| model-00044-of-00065.safetensors | Weights | 7.4 GB | 1df754d60b25 |
| model-00045-of-00065.safetensors | Weights | 7.4 GB | 28b4d71e4585 |
| model-00046-of-00065.safetensors | Weights | 7.4 GB | 01ae58298937 |
| model-00047-of-00065.safetensors | Weights | 7.4 GB | 698444433593 |
| model-00048-of-00065.safetensors | Weights | 7.4 GB | d6c0e7cb9682 |
| model-00049-of-00065.safetensors | Weights | 7.4 GB | c5873606119e |
| model-00050-of-00065.safetensors | Weights | 7.4 GB | aa499f8d821b |
| model-00051-of-00065.safetensors | Weights | 7.4 GB | b7a58844143e |
| model-00052-of-00065.safetensors | Weights | 7.4 GB | 1b3c2f43a78c |
| model-00053-of-00065.safetensors | Weights | 7.4 GB | 12befe11ea39 |
| model-00054-of-00065.safetensors | Weights | 7.5 GB | d2e23857b09f |
| model-00055-of-00065.safetensors | Weights | 7.4 GB | 759254255dec |
| model-00056-of-00065.safetensors | Weights | 7.4 GB | 452562db8eb3 |
| model-00057-of-00065.safetensors | Weights | 7.4 GB | a507a568b439 |
| model-00058-of-00065.safetensors | Weights | 7.4 GB | 502d6fce0fdc |
| model-00059-of-00065.safetensors | Weights | 7.4 GB | f066e8339eb7 |
| model-00060-of-00065.safetensors | Weights | 7.5 GB | 7263677b99d1 |
| model-00061-of-00065.safetensors | Weights | 7.4 GB | 0f05839bcd30 |
| model-00062-of-00065.safetensors | Weights | 26.3 GB | 0fc19d08e468 |
| model-00063-of-00065.safetensors | Weights | 553.7 MB | 0463fbfe1406 |
| model-00064-of-00065.safetensors | Weights | 7.4 GB | 5280edb91e34 |
| model-00065-of-00065.safetensors | Weights | 3.8 GB | 2917cab2f73f |
| config.json | Configuration | 425.3 KB | — |
| generation_config.json | Configuration | 200 B | — |
| model.safetensors.index.json | Configuration | 10.7 MB | efd3d28c35b1 |
| processor_config.json | Configuration | 1.4 KB | — |
| quantization_config.json | Configuration | 390.0 KB | — |
| special_tokens_map.json | Configuration | 114 B | — |
| LICENSE.md | Documentation | 8.3 KB | — |
| README.md | Documentation | 3.0 KB | — |
| USE_POLICY.md | Documentation | 4.9 KB | — |
| chat_template.jinja | Other | 16.3 KB | — |
| l4_200k_base | Other | 3.6 MB | — |
| .gitattributes | Repository | 1.6 KB | — |
| tokenizer.json | Tokenizer | 28.1 MB | ac2f41ef2444 |
| tokenizer_config.json | Tokenizer | 364.7 KB | — |
License and Download
- License
- other
- Access
- Open weights, no gate
- Download size
- 508.9 GB
Download from Intel AI
Released by Intel AI through its official repository on Hugging Face.
Built From
- Derived from meta-models/Muse-Spark-Open-816B-A42B
- Described by arXiv:2309.05516
- Quantized from meta-models/Muse-Spark-Open-816B-A42B
Memory Requirements
| Precision | Weights in memory |
|---|---|
| As published | 508.9 GB |
| 16-bit | 1644.1 GB |
| 8-bit | 822.0 GB |
| 4-bit | 411.0 GB |
Weights only, from the published parameter count; the key-value cache and runtime add to this.
Questions About Muse-Spark-Open-816B-A42B-MXFP4
How much GPU memory does Muse-Spark-Open-816B-A42B-MXFP4 need?
About 1972.9 GB at 16-bit and 493.2 GB at 4-bit: the weights (822B parameters) plus a working margin. A long context needs more.
What is the cheapest GPU to run Muse-Spark-Open-816B-A42B-MXFP4 on?
At 16-bit, 8x MI325X from $16.00 an hour; at 4-bit, 2x MI325X from $4.00 an hour, at the lowest on-demand prices the SAVRN Index lists.
What license is Muse-Spark-Open-816B-A42B-MXFP4 released under?
other, as its publisher declares it. Read the license text before commercial use.
What is Muse-Spark-Open-816B-A42B-MXFP4's context length?
1,048,576 tokens, from the maximum position embeddings in its published configuration.
Similar Models
Join our WeChat or Discord community. Check out the GLM-5 technical blog. Use GLM-5 API services on Z.ai API Platform. One click to GLM-5. [ Paper ] [ GitHub ] We are launching GLM-5, targeting complex systems engineering and long-horizon agentic tasks. Scaling is still one of the most important ways to improve the intelligence efficiency of Artificial General Intelligence (AGI). Compared to GLM-4.5, GLM-5 scales from 355B parameters (32B active) to 744B parameters (40B active), and increases pre-training data from 23T to 28.5T tokens. GLM-5 also integrates DeepSeek Sparse Attention (DSA), largely reducing deployment cost while preserving long-context capacity. Reinforcement learning aims…
GLM-5.3 uses the same base model as GLM-5.2 — every gain comes from post-training. Compared with GLM-5.2, it is much better at complex coding and long-horizon tasks: GLM-5.3 supports deployment with the following frameworks. Feel free to try them out: - SGLang — see cookbook - vLLM — see recipes - TokenSpeed — see here - Transformers — see transformers docs - KTransformers — see tutorial - Unsloth — see guide - For deployment on the Ascend NPU platform, inference frameworks such as vLLM-Ascend, xLLM and SGLang are supported — see here. - GLM-5.3 supports controlling the thinking budget through the reasoningeffort parameter, which accepts three levels: low, high, and max. It defaults to max…
Join our WeChat or Discord community. Check out the GLM-5.2 blog and GLM-5 Technical report. Use GLM-5.2 API services on Z.ai API Platform. Try GLM-5.2 here. [ Paper ] [ GitHub ] We're introducing GLM-5.2, our latest flagship model for long-horizon tasks. It marks a substantial leap in long-horizon task capability over its predecessor GLM-5.1 and, for the first time, delivers that capability on a solid 1M-token context. GLM-5.2's new capabilities include: GLM-5.2 supports deployment with the following frameworks. Feel free to try them out: - SGLang (v0.5.13.post1+) — see cookbook - vLLM (v0.23.0+) — see recipes - Transformers (v0.5.12+) — see transformers docs - KTransformers (v0.5.12+)…
Join our WeChat or Discord community. Check out the GLM-5.2 blog and GLM-5 Technical report. Use GLM-5.2 API services on Z.ai API Platform. Try GLM-5.2 here. [ Paper ] [ GitHub ] We're introducing GLM-5.2, our latest flagship model for long-horizon tasks. It marks a substantial leap in long-horizon task capability over its predecessor GLM-5.1 and, for the first time, delivers that capability on a solid 1M-token context. GLM-5.2's new capabilities include: GLM-5.2 supports deployment with the following frameworks. Feel free to try them out: - SGLang (v0.5.13.post1+) — see cookbook - vLLM (v0.23.0+) — see recipes - Transformers (v0.5.12+) — see transformers docs - KTransformers (v0.5.12+)…
GLM-5.3 uses the same base model as GLM-5.2 — every gain comes from post-training. Compared with GLM-5.2, it is much better at complex coding and long-horizon tasks: GLM-5.3 supports deployment with the following frameworks. Feel free to try them out: - SGLang — see cookbook - vLLM — see recipes - TokenSpeed — see here - Transformers — see transformers docs - KTransformers — see tutorial - Unsloth — see guide - For deployment on the Ascend NPU platform, inference frameworks such as vLLM-Ascend, xLLM and SGLang are supported — see here. - GLM-5.3 supports controlling the thinking budget through the reasoningeffort parameter, which accepts three levels: low, high, and max. It defaults to max…