● Covered by 7 sources · 7 reportsMedium impact6 neutral1 positive
Meta Releases Muse Spark Open, an 816B Parameter AI Model Optimized for Apple Silicon
Key points
- Meta released Muse Spark Open, an 816B parameter open-weight AI model with 42B active parameters.
- Meta claims it is the "most capable US open-weight model to date" and outperforms some Chinese models.
- The model is optimized for Apple Silicon and can run locally on two M5 Ultra Mac Studios with 1TB of unified memory.
- It features a 1 million token window and is designed for coding, agentic tasks, and visual reasoning.
- Muse Spark Open is available via APIs and for direct download, under a license with limited commercial terms.
- Muse Spark Open is a natively multimodal model.
- It is the first model of its class to run on a desktop.
- The model is available via API platforms Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter, and Together AI.
- Model weights are downloadable from Amazon SageMaker AI, Google Cloud Model Garden, Hugging Face, and Microsoft Foundry Models.
- It runs on inference engines like SGLang, vLLM, TensorRT-LLM, Hugging Face Transformers, Apple MLX, llama.cpp, and Unsloth.
- Optimizations exist for AMD, Apple, Intel, and NVIDIA silicon.
- Meta's chief of AI Alexandr Wang announced Muse Spark Open on X.
- Meta claims Muse Spark Open competes with Alibaba's Qwen, Moonshot AI's Kimi K3, and Z.ai's GLM models.
- Meta integrated strong safeguards into Muse Spark Open.
- Meta and Apple have an unusual partnership for Muse Spark Open optimization.
Meta Launches Muse Spark Open
Meta has introduced Muse Spark Open, an 816-billion-parameter open-weight AI model with 42 billion active parameters. Meta states this model is the "most capable US open-weight model to date." It is designed for tasks such as coding, agentic work, and visual reasoning, excelling at long-horizon tasks and complex reasoning.
The model features an industry-standard 1 million token window and, according to Meta's benchmarks, performs comparably to or surpasses some leading Chinese models like GLM 5.3 and Kimi K3 in agentic tasks and visual reasoning.
Optimized for Apple Silicon
Muse Spark Open is the first model of its class optimized to run locally on Apple's M5 Ultra-equipped Mac Studio machines. Running the model locally requires a significant amount of memory, specifically 1TB of RAM. This can be achieved by pairing two Mac Studios, as Apple's unified memory architecture allows each M5 Ultra to use up to 512GB, effectively creating a supercomputer with a terabyte of unified memory.
This optimization for Apple Silicon is the result of a collaboration between Meta and Apple, enabling developers to run the model directly on their desktops. This local execution offers benefits such as enhanced security and cost savings by reducing reliance on cloud processing.
Availability and Licensing
Muse Spark Open is available through APIs from providers including Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter, and Together AI. Developers can also download the model weights directly from platforms such as Amazon SageMaker AI, Google Cloud Model Garden, Hugging Face, and Microsoft’s Foundry Model Catalog.
The model is released under the "Muse Model License," which Meta describes as having "limited commercial terms." This license allows many developers to use the model for free, specifically excluding companies with more than $20 million in gross revenue during any consecutive twelve-month period.
Applications and Impact
The model is built for complex coding tasks and agentic labor. Meta has deployed it in various agentic workflows, including coding fixes for a Mastodon client, which involved bug fixes, new features, and full builds. It has also been used in conjunction with Muse Code, Meta's AI coding terminal.
The collaboration between Meta and Apple on this project signifies a shift in historical alliances within Silicon Valley, as both companies publicly supported each other's work on Muse Spark Open.
Updates
- Meta's chief of AI Alexandr Wang announced Muse Spark Open on X.
- Meta claims Muse Spark Open competes with Alibaba's Qwen, Moonshot AI's Kimi K3, and Z.ai's GLM models.
- Meta integrated strong safeguards into Muse Spark Open.
- Meta and Apple have an unusual partnership for Muse Spark Open optimization.
- Muse Spark Open is a natively multimodal model.
- It is the first model of its class to run on a desktop.
- The model is available via API platforms Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter, and Together AI.
- Model weights are downloadable from Amazon SageMaker AI, Google Cloud Model Garden, Hugging Face, and Microsoft Foundry Models.
- It runs on inference engines like SGLang, vLLM, TensorRT-LLM, Hugging Face Transformers, Apple MLX, llama.cpp, and Unsloth.
- Optimizations exist for AMD, Apple, Intel, and NVIDIA silicon.
✨ This summary was generated by AI from the outlets' reporting listed below. It is not independently verified and may contain errors — check the original sources. How BrevFeed works →
The daily brief
One email each morning: the day's tech stories, clustered across outlets and summarized. No account needed.
One email a day. Unsubscribe in one click, any time.
Today's brief
Spend a few minutes, get the whole day. Every topic's top stories in one hands-free rundown — listen, watch, or read the transcript.
▶ Play today's brief
New every morning, and the back catalogue is archived by date.
Primary sources
GitHub ml-explore/mlx-lmHow outlets covered it
Meta launched Muse Spark Open, a new open-weight AI model designed for coding and complex reasoning tasks. Meta states the model rivals leading Chinese open-weight models and is optimized for Apple's M5 Ultra chips, marking a shift in the open-weight AI landscape.
Meta launched Muse Spark Open, an 816 billion parameter open-weight AI model with a 1 million token context window. This model is designed for complex coding and agentic tasks, and is optimized to run locally on Apple Macs, requiring at least 1 terabyte of RAM.
Meta released Muse Spark Open, an 816-billion-parameter multimodal AI model, which is optimized to run on Apple M5 Ultra Mac Studios using the MLX framework. This collaboration between Meta and Apple enables efficient on-device AI processing for complex tasks.
Meta and Apple collaborated to enable Meta's Muse Spark Open, an 816-billion-parameter AI model, to run locally on Apple Mac Studio computers. This allows developers to use the open-weight model for tasks like coding and video analysis directly on their desktops, offering security and cost benefits by avoiding cloud processing.
Muse Spark Open, an 816B parameter natively multimodal AI model with a 1M-token context window, has been released as an open-weight model. It is optimized to run on two Apple M5 Ultra Mac Studios, making it the first model of its class to operate on a desktop, and is also available via various API platforms and model hubs.
Meta has released Muse Spark Open, an 800-billion parameter open-weight AI model optimized for Apple Silicon, which requires 1TB of RAM to run locally. Meta states it is the most capable US open-weight model to date, outperforming some Chinese models in benchmarks. This model is designed to run on two M5 Ultra-equipped Mac Studios, offering a local alternative for organizations with high compute needs.
Meta launched Muse Spark Open, an 816B parameter open-weight model with 42B active parameters, which it claims is the most capable US open-weight model to date. The model is available via APIs and for direct download, and is designed for coding and agentic work, excelling at long-horizon tasks and complex reasoning.