Integration guides · 2026-09-27
Can you download Muse Spark 1.2? Weights, clients and API access
Separate Muse Spark 1.2 model weights from client downloads and API access. Verify the official distribution before assuming a package runs the model locally.
Begin with the access that is verified
On 27 September 2026, Meta's model documentation describes Muse Spark 1.2 as an API offering. This research did not verify an official downloadable weight package for Muse Spark 1.2. Providing a supposed file URL or local installation command would therefore be misleading. That is not a claim that no future version can be released; it separates verified access from an expected release.
Meta's earlier statement referring to an open-weight release does not, by itself, establish that files have been published. Nor does another downloadable Muse model establish availability for 1.2. Check the exact model, distributor and contents instead of relying on similar names in search results.
Identify what you are downloading
A terminal client, desktop application or SDK may be a small download while inference still takes place on a remote API. Model weights are the files a runtime loads to perform local inference. An interface opening offline does not establish that the model can answer without an internet connection. Read which component the installation actually provides.
When a package description mentions a model, ask whether it contains weights or only client code, whether an account is required and which model identifier it calls. A filename containing Spark is not verification. Record client and model versions separately so that an application update does not obscure which model your workflow used.
| Download | What it provides | What it does not establish |
|---|---|---|
| SDK or client | Software communicating with an API | Local model execution |
| Desktop interface | User interface | Offline inference |
| Model weights | Files loaded by an inference runtime | Hardware compatibility or license conditions |
Follow the official release chain
To investigate a weight-release claim, start at the model producer's official page. Look for a consistent path from that page to the repository, release notes, actual files and license. A third-party repository with a similar name is not automatically an official distribution. Do not treat an unlinked package as authentic just because its description sounds convincing.
Even when weights are available, read runtime requirements, format and hardware conditions separately. A command copied from an old screenshot may install something different today. The date of this guide is a reference point, not a substitute for checking sources on the day you install. Leaving the decision open is better than spending time evaluating the wrong model.
Choose the correct identifier for API use
On LLMTR, the Standard row is meta/muse-spark-1.2 and the Contributor row is meta/muse-spark-1.2-contributor. They are request identifiers, not download addresses. An application calling the model should handle credentials securely and obtain its supported request format from the relevant integration guide.
Keep the first trial to a short, nonconfidential passage. A successful HTTP response does not establish that every parameter behaved as expected or that the result fits your task. Inspect the answer and record the selected model identifier. Completing client installation and completing model evaluation are separate milestones; do not combine them into one checkbox or assume that a working connection proves application quality.
Contributor is not a local model license
Contributor describes a separate pricing and data arrangement. Meta offers that tier in exchange for permission to use prompts and completions for training. Consider those conditions before starting any trial involving confidential, personal or customer information. Access options surrounding the same model name do not necessarily have identical data conditions.
Choosing Standard does not, by itself, establish local execution, zero retention or processing in a particular country. Verify each required condition independently. This article does not choose a contract for you; it prevents Contributor from being mistaken for open source, permission to download weights or the ability to run offline. Keep the data decision separate from the convenience of installing a familiar client.
Keep a local-execution requirement explicit
If your task must work without the internet or process data in an environment you control, installing an API client does not satisfy that requirement. Evaluate another model with verified official weights and licensing as a separate candidate. Do not claim that the alternative has Muse Spark 1.2's quality or behavior; evaluate representative tasks.
Write three lines in your decision record: required execution environment, verified access method and the condition still unmet. If API access is suitable, continue with the model guide. If downloadable weights are mandatory, do not turn an unverified link into an installation recommendation. The search for a download should end with a clear distinction between what exists, what has been verified and what your application actually needs.
Frequently asked questions
Is there a Muse Spark 1.2 weight download in this guide?
No. The research on 27 September 2026 did not verify an official weight package. Documented API access is distinguished from a possible future release.
Does installing a Muse client run the model locally?
Not by itself. A client may connect to a remote API. Check where the weights and inference actually reside.
Is Contributor an open-source edition?
No. Contributor is an API tier with training-related data conditions. It does not mean a local model license or downloadable weights.