Meta released Muse Spark Open on October 15, 2026, under the headline “The Most Capable US Open-Weight Model to Date.” It is natively multimodal, has 816 billion total parameters with 42 billion active, a 1-million-token context window, and is built for coding and agentic work.
The weights are on Hugging Face, Amazon SageMaker AI, Google Cloud Model Garden and Microsoft Foundry Models, and hosted APIs are offered through Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter and Together AI. It ships under the Muse Model License, which Meta describes as having “limited commercial terms, which allow many developers to use the model for free.” Meta says it is the first model in its class to run on a desktop, using two Apple M5 Ultra Mac Studios with MLX, and it released SGLang kernels it says give 1.03x to 3.95x kernel-level speedups.
Meta says the model received additional safety post-training and “met the thresholds our framework requires for release” under its Superintelligence Scaling Framework, with a full evaluation report to come.
Why it matters: when Muse Spark launched in April 2026 it was Meta’s first model without open weights, a break with the Llama strategy. Muse Spark Open returns Meta to large open-weight releases, three days after it launched the closed Muse Wonder, settling into a split where the top tier stays closed and a strong tier ships open.
What it does not show: the post gives no benchmark scores in text, so the “most capable US open-weight” claim rests on Meta’s own evaluation report, and the Muse Model License is not an open-source license.