IT之家 · 智能时代 · 10/16/2026, 07:36:51
Meta Releases Muse Spark Open: 816B MoE Model with 1M Context Runs Locally on Dual M5 Ultra Macs
Meta has released Muse Spark Open, a native multimodal model with 816B total parameters (42B active) and a 1M-token context window, claiming it as the strongest US open-weight model. Optimized for coding and agents, it uniquely supports local execution on just two Apple M5 Ultra Mac Studios and is available via major cloud APIs and inference engines.
SOURCE COVERAGEFull Original Report
On October 16, it was reported that on October 15 (local time), Meta released the open-weight AI model Muse Spark Open. Meta officially claims that Muse Spark Open is the most capable US open-weight model to date.
Muse Spark Open features a native multimodal architecture with a total parameter count of 816B (42B active parameters) and supports a context window of 1M tokens. Designed specifically for code generation and agentic collaboration, the model demonstrates exceptional performance in handling long-horizon tasks, complex software engineering, and multi-step reasoning.
Thanks to the latest MLX framework, Muse Spark Open is the first model in its class capable of running locally on desktops using just two Apple M5 Ultra chips (512 GB memory version) in a Mac Studio.
Starting today, developers can access Muse Spark Open via API platforms such as Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter, and Together AI. Model weights are also available for public download on Amazon SageMaker AI, Google Cloud Model Garden, Hugging Face, and Microsoft Foundry Models.
The model has been adapted for mainstream inference engines including SGLang, vLLM, TensorRT-LLM, Hugging Face Transformers, Apple MLX, llama.cpp, and Unsloth, with deep optimizations completed for AMD, Apple, Intel, and NVIDIA chips. Developers can also easily perform secondary development and customization based on the PyTorch framework.
During the post-training phase, Muse Spark Open underwent targeted enhancements for safety alignment. Benchmark results indicate that the model provides robust safety guarantees across all tested risk dimensions.
The model weights are released under the Muse Model License, which includes restrictive commercial terms; however, the vast majority of developers can obtain and use the model for free.
IT Home provides the following relevant test data:
