Text and chat · Meta

Muse

Meta’s text models: Muse Wonder is its first frontier model, reading text, images, audio and video over 1M tokens, and Muse Spark Open is the open-weight one.

Anonymized, some versions private High cost
Latest versionMuse Spark OpenOctober 15, 2026
Artificial Analysis Intelligence Index52 points#7 of 90 on Artificial Analysis Intelligence Index
Price$11.24 per million tokens written$0.00787 a page written
Key spec1M tokens of context

What sets it apart

Muse is Meta’s current model line, separate from the older Llama family. Muse Wonder is a closed frontier model for long agentic work, and Muse Spark Open is a smaller open-weight sibling released three days later.

  • 1M-token context on both models
  • Wonder reads images, video and audio
  • Spark Open ships open weights, 816B with 42B active

Strengths and limits

Where it shines

  • Muse Wonder scores 52 on the Artificial Analysis Intelligence Index and Muse Spark Open 39, the latter ranked among the open-weight models there.
  • Meta presents Wonder as built to plan, recover from errors and keep a multi-step task on course, with an optional maximum reasoning effort.
  • Both models accept images, up to 10 per request, and call tools.

Think twice when

  • Muse Wonder is proprietary: no weights to download. Spark Open’s Muse Model License asks for a “Built with Muse Spark Open” credit and a separate agreement for large model-as-a-service resellers.
  • Spark Open reads images but not video or audio in this catalog; only Wonder takes them.
  • Muse Wonder was released on 12 October 2026, Muse Spark Open on 15 October 2026.
  • Both accept up to 1,048,576 tokens of context and write up to 131,072 tokens in one reply.

Use cases

Work across a video and its documents

Watch this recorded demo, read the attached spec, and list every place where the product behaves differently from the spec.

Muse Wonder takes video, audio, images and PDFs in one request.

Long coding task with tools

Find why the nightly export job times out, propose a fix, and write the test that would have caught it.

Both models are tuned for tool calling and multi-step agent work.

Against its rivals

  • ClaudeClaude Opus 5.5 scores 58 on the index against 52 for Muse Wonder.#1 of 90 on Artificial Analysis Intelligence Index (58 points)
  • GPTGPT-6 Astra scores 53 on the index, one point above Muse Wonder.#4 of 90 on Artificial Analysis Intelligence Index (53 points)
  • GPT OSSAnother open-weight family from a large lab, to compare with Muse Spark Open.#79 of 90 on Artificial Analysis Intelligence Index (12 points)
Compare side by side ↗

Version history

Each version, dated, with what it changed over the one before.

Artificial Analysis Intelligence Index, version after versionHigher is better · Artificial Analysis

Hollow points are earlier versions no longer offered in Sub Rosa.

See the numbers
VersionReleasedpoints
Muse Wonder2026-10-1252
Muse Spark Open2026-10-1539
  1. Muse Wonder
    • Meta’s first frontier model, for personal and professional agentic tasks, released with a larger Muse Wonder Full sold only to Muse Code subscribers
    • Available through the Meta Model API and Muse Code
    Source : about.fb.com
  2. Muse Spark OpenLatest
    • Open-weight sparse mixture-of-experts reasoning model, 816B parameters with 42B active, for coding and agent work
    • Released under the Muse Model License
    Source : artificialanalysis.ai

Benchmarks

How to read these scores: what a benchmark tells you.

Artificial Analysis Intelligence Index: #7 of 90 models measured in the catalog. One score averaging about ten independent tests of reasoning, knowledge, math, coding and agent work, all run the same way for every model. Higher is better. Each model is listed at its highest reasoning effort. Two or three points apart is a close call; ten points is a different class of model.

Artificial Analysis Intelligence IndexHigher is better · Artificial Analysis, October 16, 2026
  1. 1. Claude Opus 5.5Claude58
  2. 2. Claude Sonnet 5.5Claude56
  3. 3. Claude Fable 5.1Claude53
  4. 4. GPT-6 AstraGPT53
  5. 5. Gemini 4 ArgonGemini53
  6. 6. GPT-6.1 SolGPT52
  7. 7. Muse WonderMuse52
  8. 8. Claude Opus 5Claude51
  9. 9. Claude Fable 5Claude50
  10. 10. GPT-6 SolGPT48
  11. 28. Muse Spark OpenMuse39
See the numbers
Modelpoints
1. Claude Opus 5.558
2. Claude Sonnet 5.556
3. Claude Fable 5.153
4. GPT-6 Astra53
5. Gemini 4 Argon53
6. GPT-6.1 Sol52
7. Muse Wonder52
8. Claude Opus 551
9. Claude Fable 550
10. GPT-6 Sol48
28. Muse Spark Open39

All 2 models in this family

ModelUsePrivacyPrice
Muse Wondermuse-wonderTextAnonymized$11.24 per million tokens written$0.00787 a page written
Muse Spark Openmuse-spark-openTextPrivate$2.389 per million tokens written$0.00167 a page written

Specifications

SpecificationMuse WonderMuse Spark Open
ReleasedOctober 12, 2026October 15, 2026
ParametersNot disclosed816B total, 42B active
WeightsClosedOpen, Muse Model License
Acceptstext, images, video, audiotext, images
Producestexttext
Context1,049K tokens1,049K tokens
Longest answer131K tokens131K tokens
AbilitiesReads images, Uses Sub Rosa’s tools, Thinks before answering, Tuned for codeReads images, Uses Sub Rosa’s tools, Thinks before answering, Tuned for code
Price$11.24 per million tokens written$2.389 per million tokens written
PrivacyAnonymizedPrivate

Prices, context and lengths from the live catalog on October 16, 2026; parameters and licenses from the maker.