IT之家 October 16 news — On October 15 local time, Meta released the open-weights AI model Muse Spark Open. Meta officially claims thatMuse Spark Open is the most capable American open-weights model to date。
Muse Spark Open adopts a native multimodal architecture, with 816B total parameters (of which 42B are activated parameters), and supports a 1M token context window. The model is designed for code building and agent collaboration, demonstrating outstanding performance in handling long-horizon tasks, complex software engineering, and multi-step reasoning.
Thanks to the latest MLX framework, Muse Spark Open becomes the first model in its class that can run locally on the desktop with just two Mac Studio machines equipped with Apple M5 Ultra chips (512 GB memory version).
Starting today, developers can call Muse Spark Open through API platforms such as Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter, and Together AI; the model weights have also been simultaneously made available for public download on Amazon SageMaker AI, Google Cloud Model Garden,Hugging Face and Microsoft Foundry Models.
The model has been adapted to mainstream inference engines such as SGLang, vLLM, TensorRT-LLM, Hugging Face Transformers, Apple MLX, llama.cpp, and Unsloth, and has been deeply optimized for AMD, Apple, Intel, and NVIDIA chips. Developers can also conveniently perform secondary development and customization based on the PyTorch framework.
Muse Spark Open received targeted enhancements in safety alignment during the post-training stage. Benchmark results show that the model has robust safety safeguards across all tested risk dimensions.
The model weights are released under the Muse Model License with restrictive commercial terms, and the vast majority of developers can obtain and use the model for free.
IT之家 attaches the relevant test data below:
Advertising notice: This article contains external jump links (including but not limited to hyperlinks, QR codes, passcodes, and other forms), which are used to convey more information and save screening time. The results are for reference only. All IT之家 articles containing external links include this notice.