Meta AI Research on October 15, 2026 released the weights for Muse Spark Open, describing it as the most capable U.S. open-weight model to date. The model is natively multimodal, with 816 billion total parameters (42 billion active) and a 1 million-token context window, and is positioned for coding, long-horizon agent work, and multi-step reasoning.
According to Meta’s announcement, Muse Spark Open is the first model in its class designed to run on the desktop using two Apple M5 Ultra Mac Studios with the latest MLX framework. Apple’s unified memory allows up to 512 GB per Mac Studio; Meta says two machines can hold the full model and behave like one system with a terabyte of unified memory. Independent coverage, including Engadget and TechCrunch, notes that local runs still demand roughly 1 TB of RAM-class memory.
Hosted inference is available immediately through API platforms including Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter, and Together AI. Weights are listed for download on Amazon SageMaker AI, Google Cloud Model Garden, Hugging Face, and Microsoft Foundry Models. Meta also cites support across SGLang, vLLM, TensorRT-LLM, Hugging Face Transformers, Apple MLX, llama.cpp, and Unsloth, with PyTorch for further customization.
Meta says it tested on-device agent workflows with Muse Code powered by Muse Spark Open, including shipping iOS changes to the open-source Mastodon client Ice Cubes and driving DaVinci Resolve end to end for footage review and vertical cropping. The company is releasing the weights under the Muse Model License with limited commercial terms, and states that Muse Spark Open received additional safety post-training and met thresholds in Meta’s Superintelligence Scaling Framework for release.
DigiEditorial verified the launch against Meta’s research blog and independent reports from TechCrunch, Engadget, SiliconANGLE, and The Verge. Benchmark figures cited by Meta are company-published; third-party leaderboards were not independently re-run for this report.