AI News
Meta Releases Muse Spark Open: 816B-Parameter Open-Weight Model for Coding and Agents
Meta announced Muse Spark Open on October 15, 2026, an open-weight model aimed at coding and agentic tasks. The model carries 816 billion total parameters with 42 billion active at a time, and supports a 1 million-token context window, putting it among the largest open models Meta has released for developers and researchers.
The release is aimed squarely at the growing market for autonomous AI agents. A 1M-token context lets the model hold entire codebases, long documents, and extended agent trajectories in memory, which matters for the multistep workflows that coding assistants and task-running agents increasingly rely on.
Open weights, downloadable and hosted
The model weights are available for download from Hugging Face, and the model is also served through a set of API providers: Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter, and Together AI. That spread of hosting options means developers can either run the model themselves or access it through cloud APIs without depending on a single vendor.
Notably, Meta says the model can run locally on two Apple Mac Studio machines with the M5 Ultra chip. That is a striking claim for a model of this size: frontier-class open weights that fit into desktop-scale hardware rather than data-center clusters. For independent developers and small teams, local deployment means lower costs, more privacy, and no per-token billing.
Meta's framing: AI shouldn't be concentrated
Meta's CEO framed the release around openness, arguing that advanced AI shouldn't be concentrated in a few companies. The open-weights strategy is consistent with Meta's long-running approach of releasing powerful models publicly while monetizing elsewhere, in contrast to rivals that keep their flagship models closed behind APIs.
The release lands in a crowded field. With Google, OpenAI, and Anthropic all racing on agent capabilities, an 816B-parameter open model for coding and agents gives the open-source ecosystem a new heavyweight to build on — and gives developers an alternative to locked-in proprietary APIs. Developers can fine-tune or distill the model on their own terms, and enterprises can audit exactly what they are deploying, two advantages that closed models by definition cannot offer.