IonBin

Independent reviews. IonBin has no partnerships with the tools on this page and was not paid for these listings. Our evaluations are our own, based on public documentation and our research. Product names are trademarks of their owners and are used for identification only. How disclosure works.

Model · Meta

Muse models (Meta)

Verified Oct 17, 2026 Review by Nov 16, 2026 4 sources
Short answer

Meta's developer site, llama.com, now leads with the Muse model family. Muse Wonder targets coding, long-running agents and computer use. Muse Spark is a cheaper agentic model, and Muse Spark Open is an 816B open-weight model under the Muse Model License. Hosted models are served through the Meta Model API, OpenRouter and Muse Code.

What it is

Muse is the model family Meta currently promotes to developers on llama.com. It includes hosted models served through Meta’s own API and Muse Spark Open, which Meta releases with open weights you can run on your own infrastructure. The site still links to Llama documentation and repositories for Meta’s earlier open models.

How it works

Hosted models are called through the Meta Model API. Meta says it works with existing OpenAI-compatible SDKs, so a team can swap a base URL rather than rewrite code. The models are also listed on OpenRouter. Muse Code is Meta’s terminal tool for multi-agent coding with these models. Muse Spark Open can be downloaded and served with SGLang, vLLM or Apple MLX, or used through hosted providers.

Specs and pricing (checked 2026-10-17). Meta’s API page lists a 1M-token context window for its hosted models. Muse Wonder costs $4 per million input tokens and $20 per million output tokens. Muse Spark 1.3 costs $1.25 and $4.25. Cheaper “contributor” versions are listed at $0.10 and $0.20, but Meta says they’re used to improve its products. Pick the standard version for business data.

Benefits

  • Self-hosting path. Open weights can keep data on your own infrastructure, which matters in regulated work.
  • Low switching cost. OpenAI-SDK compatibility means existing integrations can try it quickly.
  • Agent focus. Meta positions Wonder for long multi-step tasks and computer use.

Limits and risks

  • License first. Muse Spark Open uses Meta’s own Muse Model License. “Open-weight” isn’t the same as open source, so read the license before any commercial use.
  • Watch the data terms. The contributor versions trade a lower price for Meta using your traffic to improve its products.
  • Vendor claims. Meta describes Muse Spark Open as the most capable US open-weight model and cites benchmark wins for Wonder Full. These are Meta’s claims. Test on your own tasks.
  • Self-hosting is a job. An 816B model needs serious hardware, plus monitoring and security work that a hosted API handles for you.

Checked specs

Snapshot from official sources on Oct 17, 2026. Model lineups and prices change often; this page is re-verified every 30 days.

Headline modelsMuse Wonder, Muse Wonder Full, Muse Spark 1.3, Muse Spark Open [1]
Context window (hosted models)1M tokens [2]
Muse Wonder price per 1M tokens (checked 2026-10-17)$4.00 input, $0.20 cached input, $20.00 output [2]
Muse Spark 1.3 price per 1M tokens (checked 2026-10-17)$1.25 input, $0.15 cached input, $4.25 output; Meta lists it as not used to improve its products [2]
Contributor variantsmuse-spark-1.3-contributor at $0.10 input / $0.20 output, which Meta says is used to improve its products [2]
Muse Wonder FullSubscription access through Muse Code only [2]
Muse Spark Open816B open-weight model; text, image, video and audio input; 1M context; Muse Model License; released October 15, 2026 [4]
AccessMeta Model API (works with OpenAI-compatible SDKs, per Meta); OpenRouter; Muse Code [1]

Best-fit stacks

Stack 08

Dev, web & IT

Design, code, host and ship sites and internal tools, with an AI editor doing most of the typing.

Agent compatibility

API accessYesOfficial docs ↗
AppsMuse Code (terminal coding tool)
Also offered viaOpenRouter; Muse Spark Open on several inference hosts

Levels come from each tool's recorded, sourced compatibility. How the levels work.

Sources

  1. llama.com home page · Metaofficial accessed Oct 17, 2026
  2. Meta Model API · Metaofficial accessed Oct 17, 2026
  3. Muse Wonder · Metaofficial accessed Oct 17, 2026
  4. Build with Muse Spark Open · Meta, 2026-10-15official accessed Oct 17, 2026

Practitioner views are restated in our words and labeled as views, not measured results. Product facts come from official sources dated above.

Suggest an edit

IonBin's wiki isn't user-edited. Corrections go to our Directory Curator, who checks them against primary sources before anything changes. Spot something wrong or out of date on “Muse models (Meta)”? Email us with a link to a primary source.

Suggest an edit by email or write to [email protected]