Back to News

Meta releases Muse Spark Open, an 816B-parameter open-weight multimodal model

#open-weight#multimodal#apple-silicon#meta

Meta released Muse Spark Open, an open-weight multimodal mixture-of-experts model with 816B total and 42B active parameters and a 1M-token context window, optimized for Apple silicon and claimable to run on two Mac Studios. The model targets code and agentic tasks, with day-0 support from vLLM and other inference engines, and marks a collaboration between Meta and Apple.

Coverage timeline

  1. Techmeme
    Meta releases Muse Spark Open, saying it is the most capable US open-weight model, with 816B total and 42B active parameters and a 1M-token context window (Meta AI Research)

    Meta AI Research : Meta releases Muse Spark Open, saying it is the most capable US open-weight model, with 816B total and 42B active parameters and a 1M-token context window — Muse Spark Open is available today and brings leading agentic capabilities to your desktop for the first time on Apple M5 Ultra Mac Studios.

  2. Hugging Face Blog
    Welcome Muse Spark Open: open-weight frontier multimodal mixture-of-experts model from Meta
  3. vLLM BlogvLLM Team
    Day 0 support for Muse Spark Open on vLLM

    Day-0 vLLM support for Muse Spark Open: 816B MoE, 1M context, multimodal inputs, DSpark speculative decoding, NVIDIA and AMD serving, initial TPU support, and LoRA RL.

  4. TechCrunch AILucas Ropek
    Meta launches an open-weight AI model that runs on Macs

    Meta and Apple have teamed up for the launch of the social media giant's newest model, Muse Spark Open, which will run locally on Apple silicon.

  5. 机器之心机器之心
    Meta杀回开源：816B新模型，两台Mac Studio就能跑

    编辑｜泽南 没想到 Meta 重回开源的方式，是和苹果携手。 本周五凌晨，Meta 突然发布了 Muse Spark Open，号称「最强美国开源权重大模型」。这是一款原生多模态模型，参数量 816B，激活参数 42B，上下文 1M，专门为代码和智能体任务设计，擅长长期任务、复杂多步骤推理。 Meta 希望未来每个人都有能在家中运行高级模型的自由。扎克伯格说了，开源版 Muse Spark 是专门对 Mac 设备优化的，他和苹果新 CEO 约翰・特努斯（John Ternus）进行了合作。 具体来说，Muse Spark Open 是一款类似于 Muse Spark 1.3 的模型，利用苹果最新版本的 MLX 框架针对苹果芯片的统一内存架构进行了优化，因此两台 Apple M5 Ultra Mac Studio（512G）本地就能跑。 HuggingFace 链接：https://huggingface.co/meta-models/Muse-Spark-Open-816B-A42B 苹果的统一内存架构让 M5 Ultra 的集成 GPU 能够为每台 Mac Studio 使用高达 512GB 的内存，因此两台机器就能容纳一个完整的大模型。该芯片 1.2TB/s 的内存带宽足以读取每个 token 所需的小块内存。最终，两台 Mac Studio 的运行速度堪比一台拥有 1TB 统一内存的超级计算机。 不过必须要说的是，两台 512G 的 Mac Studio 价格已经超过了 4 万美元。 考虑到苹果和 Meta AI 之间过往的冷淡关系，两家公司此次合作着实令人惊讶。Meta AI 首席执行官 Alexandr Wang 在接受采访时表示，他们「很高兴看到父母再次和好如初」。Wang 解释说，苹果和 Meta AI 已经合作数月。他还特别感谢苹果前高管 Daniel Gross 在促成两家公司友好合作方面发挥了作用。 Meta 也帮开发者做好了 SGLang、vLLM、TensorRT-LLM、Hugging Face Transformers、Apple MLX、llama.cpp 和 Unsloth 等主流推理引擎的适配，并针对 AMD、苹果、英特尔和英伟达芯片进行了优化。开发者可以使用 PyTorch 框架轻松地进一步定制模型。 关键基准参数： 看起来，重走

Related stories

  • Meta and Apple collaborate to run Muse Spark Open locally on M5 Ultra Mac Studios

    Meta announced Muse Spark Open, claiming it is the most capable US open-weight model to date and the first in its class to run on desktop using two Apple M5 Ultra Mac Studios with the MLX framework. Meta and Apple, typically rivals, worked together on the launch, with Meta CEO Mark Zuckerberg and Apple CEO John Ternus publicly supporting each other.

    2 reports · 2 sources

  • Meta launches Muse Wonder and Muse Wonder Full AI models

    Meta has released two new AI models, Muse Wonder and Muse Wonder Full, claiming state-of-the-art performance on personal and professional agentic tasks. The models introduce Active Memory and Active Focus features and are available to Muse Code subscribers, with Muse Wonder also accessible via the Meta Model API. Meta plans to eventually power its Muse AI agent with Wonder, starting with subscribers.

    9 reports · 4 sources

  • Meta's Muse Appears to Use OpenAI Model Labeled 'muse-special'

    A Hacker News report suggests that Meta's Muse product appears to rely on an OpenAI model labeled 'muse-special'. The finding was reported on September 25, 2026, and has drawn attention within the AI research community.

    1 reports · 1 sources143 points · 50 comments

  • StepFun Unveils Step 5 Preview: 600B MoE, 27B Active, Open-Source Top 2

    StepFun released Step 5 Preview, a 600B-parameter MoE model with 27B active parameters and 1M context, available as an open-weight checkpoint on Hugging Face. In Artificial Analysis evaluations, it scored 44, ranking as the top open-source model after one other, with pricing at $1 per million input tokens and $2.7 per million output tokens. Hands-on tests showed strong agentic capabilities, including autonomously building 3D assets and a LEGO racing game.

    2 reports · 2 sources

News updates every hour. Subscribe in your RSS reader to get new stories automatically.

Subscribe