Model release

Muse Spark Open 816B-A42B

Muse Spark Open 816B-A42B is a model release in the Muse (Meta) family from Meta (Meta Superintelligence Labs)1 4, version Muse-Spark-Open-816B-A42B, released Oct 15, 2026. Its weights are available only on request or with approval2 3 1, under the Muse Model License (version Muse Spark Open)2 1.

Muse Spark Open is the open-weight version of Meta's Muse Spark, released on October 15, 2026. The model card describes a mixture-of-experts transformer with 816B total and 42B active parameters (256 experts, 8 active per token, 62 layers) plus a 1.8B-parameter perception encoder. It accepts text, image, video, and audio input, produces text, and supports a 1M-token context at inference. Meta says this version received additional safety post-training for use as a standalone model.1 4

Model hub: Model card (Hugging Face) (external site: huggingface.co)
Model hub: Muse Spark Open collection (BF16 and NVFP4) (external site: huggingface.co)
License: License (Muse Model License) (external site: huggingface.co)
Release notes: Meta developer blog post (external site: dev.meta.ai)

On AI on Hugging Face: meta-models/Muse-Spark-Open-816B-A42B

Last reviewedOct 18, 2026Entry updated Oct 18, 2026Documented release Oct 15, 2026

Availability and license

Overall availability

Partial

The Hugging Face repositories are gated with manual approval: downloading requires accepting the Muse Model License and submitting a form asking for name, date of birth, country, affiliation, and job title. The model card says the model is also offered through third-party inference providers.2 3 1

Availability fact review: Oct 18, 2026

Availability is separate from permission: read the license before using or redistributing.

Muse Model License (version Muse Spark Open) (external site: huggingface.co)2 1
Applies to: Everything in the release · custom license (no SPDX identifier) · fact reviewed Oct 18, 2026 · effective Oct 15, 2026

Custom Meta license (card license field "other", license_name "muse-model-license"). The LICENSE.md file is behind the access gate and was not fetched; the terms here are from the license text Meta publishes in the repository's gating prompt, dated October 15, 2026. It covers the model, code, weights, configuration, and documentation and grants broad rights to use, modify, fine-tune, distill, distribute, and sell, subject to conditions: copies must include the license and the incorporated Usage Policy (Exhibit A), and products using the model or its derivatives must prominently display "Built with Muse Spark Open". A licensee offering models to third parties as an API service, whose group gross revenue exceeds US$20 million in any 12-month period, must notify Meta and sign a separate agreement first; internal use, research, and use inside the licensee's own non-API products are exempt. Meta claims no ownership of outputs. The Usage Policy prohibits uses including military and weapons applications, and the model is not intended for people under 18.2 1

Component reuse rights

A readable or downloadable component is not automatically reusable. These indicators concern recorded license evidence, not system certification.

weights: Unknown — no complete fact-level rights review
code: Unknown — no complete fact-level rights review
data: Unknown — no complete fact-level rights review
documentation: Unknown — no complete fact-level rights review

No complete system-rights review is recorded for this release.

Model-disclosure tier

Computed from the checklist below using USASI rubric v0.2. An editorial category, not a certification.

Restricted weights

The weights can be obtained only by request, with approval, or by some users — for example a gated download that the publisher reviews. Not counted as open-weight.

How tiers are computed

Public materials checklist

Items for a model under USASI rubric v0.2. Unknown means unassessed or insufficient evidence.

Public materials checklist for Muse Spark Open 816B-A42B
Item	Status	Notes and evidence
WeightsCan the general public download the model parameters for this release?	Partial	BF16 weights (Muse-Spark-Open-816B-A42B) and an NVFP4 checkpoint (Muse-Spark-Open-816B-A42B-NVFP4) are on Hugging Face behind a manually approved access gate.2 3 1
Inference codeIs code for running the model published?	Public	The model card lists support in SGLang, vLLM, Hugging Face Transformers, TensorRT-LLM, Unsloth, llama.cpp, and MLX, and Meta's developer blog gives serving commands for several of them. No separate Meta reference inference repository was found.1 4
Training codeIs the code used to train the model published?	Unknown	Meta points to fine-tuning and reinforcement-learning tooling (including TorchTitan and Hugging Face TRL); no code used to train this model was found.
Training-data informationDoes the information cover provenance, scope, acquisition, selection, labeling, processing, and where data or alternatives can be obtained? Access alone does not establish completeness.	Partial	The model card describes text, image, audio, and video data from public sources, web crawls, third parties, Meta products and services, and synthetic generation, cleaned, deduplicated, and filtered, covering more than 100 languages. No dataset listing is given.1
Training-data accessCan the training data be obtained? This is independent of information completeness and reuse rights; original unshareable data need not be downloadable.	Unknown	Not assessed.
Complete training pipelineIs the complete base-training and preprocessing pipeline published, including configuration? Fine-tuning code or an inference SDK alone is insufficient.	Unknown	Not assessed.
Legacy data assessment (v0.1)Historical assessment combining download access and disclosure. Preserved for traceability; excluded from the v0.2 tier calculation. See the new separate assessments above.	Unknown	Not assessed.
Training recipeAre the training configuration and procedure documented in enough detail to follow?	Unknown	The sources say only that this version shares its base model with Muse Spark 1.3 and has additional safety post-training; no recipe details were found.
Evaluation materialsPublic = evaluation code or prompts that let others re-run the evaluations are published. Partial = results only.	Partial	The model card reports benchmark and safety evaluation results and links an evaluation methodology report, which was not reviewed here. Evaluation code was not located.1

What it is useful for

The model card lists long-horizon reasoning, agentic task completion, coding agents, tool use, multimodal reasoning over screenshots, documents, images, audio, and video, synthetic data generation, and LLM-as-a-judge evaluation. Meta recommends deploying it inside a larger system with guardrails suited to the use case.1

Run and use notes

Documented facts only. No hardware or performance claims are made without a cited source and stated assumptions.

The model card states that running the model locally requires at least 700GB of GPU memory for the NVFP4 checkpoint and 2TB for BF16, plus a supported inference framework. Meta's developer blog says MLX can serve the NVFP4 checkpoint across two Mac Studios with M5 Ultra and 512GB of unified memory each, or four with 256GB each.1 4
The model card recommends temperature 1.0, top_p 1.0, and top_k disabled, and documents a reasoning_effort setting (low, medium, high, or xhigh; default xhigh).1

Organization context

Meta

Provenance and derivatives

Meta's PyTorch team says Muse Spark Open shares its base model with Muse Spark 1.3, a Meta model offered without public weights; the model card says the open version adds safety post-training.5 1

Derived from: Muse Spark 1.3 — Same organization; shared base model, weights not public.

Other releases in the Muse (Meta) family

Muse Glimmer 30BModel-disclosure tier (USASI rubric v0.2): Open-weight

Muse (Meta) family overview

What this catalog does not know

Unknown means the sources reviewed for this record do not document it. It is not evidence that something does not exist.

Training code: unknown.
Training-data access: unknown.
Complete training pipeline: unknown.
Legacy data assessment (v0.1): unknown.
Training recipe: unknown.

Have a primary source? How to report a correction.

U.S. eligibility

Project eligibility rests on documented governing or maintaining entities, not on contributors.

Eligible · basis: U.S. headquarters

The model card names Meta Superintelligence Labs as author and Meta Platforms, Inc. or Meta Platforms Ireland Limited as model provider; the license names Meta Platforms, Inc. as copyright holder. Meta Platforms, Inc. is a Delaware company whose principal executive offices are in Menlo Park, California, per its Form 10-K. The Irish entity acts as licensor only for licensees in the European region.1 2 6

Assessed Oct 18, 2026

Sources

1.
Muse Spark Open Model Card (meta-models/Muse-Spark-Open-816B-A42B) (external site: huggingface.co)
Meta (Hugging Face) · Model card · published Oct 15, 2026 · accessed Oct 18, 2026
2.
meta-models/Muse-Spark-Open-816B-A42B repository metadata (license fields and gating terms) (external site: huggingface.co)
Hugging Face · Repository · accessed Oct 18, 2026
3.
Muse Spark Open collection (meta-models) repository metadata (external site: huggingface.co)
Hugging Face · Repository · accessed Oct 18, 2026
4.
Build with Muse Spark Open (external site: dev.meta.ai)
Meta · Announcement · published Oct 15, 2026 · accessed Oct 18, 2026
5.
Muse Spark Is Now an Open-Weights Model (external site: pytorch.org)
PyTorch Team at Meta · Announcement · published Oct 15, 2026 · accessed Oct 18, 2026
6.
Meta Platforms, Inc. Form 10-K for the fiscal year ended December 31, 2025 (external site: sec.gov)
Meta Platforms, Inc. (U.S. SEC filing) · Filing · published Jan 29, 2026 · accessed Oct 18, 2026

Email a correction Report a correction Edit this entry

This listing is not an endorsement, a safety assessment, or a federal approval.

Cite this page

APA style

USASI. (2026, October 18). Muse Spark Open 816B-A42B. United States of America Superintelligence. https://unitedstatesofamericasuperintelligence.com/open/muse-spark-open-816b-a42b/

BibTeX

@misc{usasi_muse_spark_open_816b_a42b,
  author = {{USASI}},
  title = {Muse Spark Open 816B-A42B},
  year = {2026},
  month = oct,
  howpublished = {\url{https://unitedstatesofamericasuperintelligence.com/open/muse-spark-open-816b-a42b/}},
  note = {United States of America Superintelligence. Last updated 2026-10-18}
}

The date is when this page was last updated. Add the date you read it if your style needs one.

For maintainers: a “Listed in USASI” badge

Links to this page. It says the entry is listed, not endorsed or rated. About the badge

Markdown (for a README)

[![Listed in the USASI catalog](https://unitedstatesofamericasuperintelligence.com/badges/listed.svg)](https://unitedstatesofamericasuperintelligence.com/open/muse-spark-open-816b-a42b/)

HTML

<a href="https://unitedstatesofamericasuperintelligence.com/open/muse-spark-open-816b-a42b/"><img src="https://unitedstatesofamericasuperintelligence.com/badges/listed.svg" alt="Listed in the USASI catalog" height="20"></a>