TellWorldAI World Archive

Home›Models

Model

Muse Spark

Muse Spark is the first model from Meta Superintelligence Labs, released on 8 April 2026 as a closed model that powers the Meta AI app and meta.ai. An open-weight sibling followed later.

Status AvailableModality Text, Image, Audio Era Generative AI for everyone

License
Proprietary
Access
API, App
Release date
Developer
Meta AI

What it is

Muse Spark is the first model from Meta Superintelligence Labs and the first in a new Muse series. Released on 8 April 2026, it powers the Meta AI app and meta.ai.[1][2]

Meta describes it as natively multimodal, with support for tool use, visual chain of thought and multi-agent orchestration. Visual chain of thought means the model reasons step by step about an image before answering. Multi-agent orchestration means several agents work on parts of a task in parallel and their results are combined. Meta also describes the model as small and fast by design, and says larger models were in development.[1][2]

Origins

In December 2025, CNBC reported that Meta's next frontier model, codenamed "Avocado", could be proprietary. That would be a shift away from the openly downloadable Llama line.[4]

Meta reported a scaling-law analysis that showed Muse Spark reaches the same capabilities as Llama 4 Maverick with over an order of magnitude less compute.[1] A scaling-law analysis studies how a model's performance changes as compute, data and size grow.

Main features

Modes. Muse Spark is offered in Instant and Thinking modes. A Contemplating mode, which runs multiple agents reasoning in parallel, was rolled out gradually on meta.ai.[2][1]

Benchmarks. Meta reported 58% on Humanity's Last Exam and 38% on FrontierScience Research in Contemplating mode.[1]

Health and evaluation. Over 1,000 physicians helped curate health training data for the model. Apollo Research found the highest rate of evaluation awareness it had observed in any model. Evaluation awareness means a model recognises that it is being tested, which can affect how its test results should be read.[1]

Access. Muse Spark is a closed model. It is free to use with a Meta account, and it is available to select partners through a private API preview. Meta said it hoped to open-source future versions.[2][3]

Rollout. From 12 May 2026, Muse Spark came to WhatsApp, Instagram, Facebook, Messenger and Threads. It also came to Ray-Ban Meta and Oakley Meta glasses in the United States and Canada.[2]

Muse Spark 1.1. On 9 July 2026, Meta released Muse Spark 1.1. It improved tool and computer use, coding and multimodal understanding. It has a 1-million-token context window and became available to developers through the Meta Model API public preview, with Replit, Cline and Box among early partners.[5]

Its place in development

Muse Spark marks a change in how Meta releases its models. The Llama line was openly downloadable, while the first Muse model was offered as a closed product first.[1][4]

An open-weight sibling, Muse Spark Open, followed on 15 October 2026.[6] Meta's stated plan to open-source future versions therefore had a first concrete step within months of the launch.[2][6]

For readers, the key point is that Muse Spark is the start of a new line rather than a continuation of Llama. The earlier Llama models are described in their own entries, and Muse Spark's place in the sequence is the step from openly downloadable models to a closed product and then back to an open sibling.[4][6]

Version history

9
  1. 2023
    • Feb 24, 2023 LLaMA

      A family of foundation language models from Meta AI, released in 7B to 65B parameter sizes and trained on public data; its 13B model was reported to beat GPT-3 on most benchmarks.

    • Jul 18, 2023 Llama 2

      Meta's family of open-weight language models released in July 2023 in 7B, 13B and 70B sizes, each with a dialogue-tuned Chat variant, issued under a custom licence that allowed free research and commercial use.

  2. 2024
    • Apr 18, 2024 Llama 3

      Llama 3 is an open model family from Meta, released on 18 April 2024 at 8 billion and 70 billion parameters. It was pretrained on over 15 trillion tokens of publicly available data.

    • Jul 23, 2024 Llama 3.1 (including 405B)

      Llama 3.1, released on 23 July 2024, added a 405-billion-parameter model to the Llama 3 family and extended its context to 128,000 tokens. Meta called the 405B the largest and most capable openly available foundation model.

    • Sep 25, 2024 Llama 3.2

      Llama 3.2, announced on 25 September 2024, added vision models at 11 billion and 90 billion parameters and small text-only models at 1 billion and 3 billion for edge and mobile devices.

    • Dec 6, 2024 Llama 3.3 70B

      Llama 3.3 70B is a text-only, instruction-tuned model Meta released on 6 December 2024. Meta said it performs close to the 405B Llama 3.1 at much lower cost and can run on a typical developer workstation.

  3. 2025
    • Apr 5, 2025 Llama 4 Scout

      Llama 4 Scout is a mixture-of-experts model Meta released on 5 April 2025. It has 109 billion total parameters, 17 billion of them active for each token, and a context window of 10 million tokens.

    • Apr 5, 2025 Llama 4 Maverick

      Llama 4 Maverick is a mixture-of-experts model Meta released alongside Scout on 5 April 2025, with 400 billion total parameters and a 1-million-token context. Its ranking on LMArena was later questioned.

  4. 2026
    • Apr 8, 2026 Muse Spark · This page

      Muse Spark is the first model from Meta Superintelligence Labs, released on 8 April 2026 as a closed model that powers the Meta AI app and meta.ai. An open-weight sibling followed later.

20262

  1. UpdateMeta releases Muse Spark 1.1 and opens the Meta Model API

    On 9 July 2026 Meta released Muse Spark 1.1, an upgrade built for agentic tasks, with a 1-million-token context and access for developers through the Meta Model API in public preview.

  2. ReleaseMeta Superintelligence Labs releases Muse Spark

    On 8 April 2026 Meta Superintelligence Labs released Muse Spark, the first model from the lab and the first of the Muse series. It is closed: its weights were not released, and it was offered through a private API preview.

Developed by1

Meta AI

Foundation for1

Muse Spark Open

Successor of1

Llama 4 Maverick

Notes and sources

  1. Introducing Muse Spark: Scaling Towards Personal Superintelligencehttps://ai.meta.com/blog/introducing-muse-spark-msl/
  2. Introducing Muse Spark: Meta's Most Powerful Model Yethttps://about.fb.com/news/2026/04/introducing-muse-spark-meta-superintelligence-labs/
  3. Meta debuts the Muse Spark model in a 'ground-up overhaul' of its AIhttps://techcrunch.com/2026/04/08/meta-debuts-the-muse-spark-model-in-a-ground-up-overhaul-of-its-ai/
  4. From Llamas to Avocados: Meta's shifting AI strategy is causing internal confusionhttps://www.cnbc.com/2025/12/09/meta-avocado-ai-strategy-issues.html
  5. Introducing Muse Spark 1.1https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/
  6. Meta launches new Muse Spark Open model, optimized to run on Mac Studiohttps://9to5mac.com/2026/10/15/meta-launches-new-muse-spark-open-model-optimized-to-run-on-mac-studio/

Cite this page: Muse Spark. TellWorld, https://www.tellworld.net/models/muse-spark. Updated 2026-11-08