Meta releases an open-weight version of Muse Spark: 816 billion parameters, aimed at coding and AI agents
On 15 October 2026 Meta published downloadable weights for its Muse Spark model, which it says has 816 billion total parameters (42 billion active) and a context window of up to 1 million tokens, according to Reuters.
Meta Platforms on Thursday, 15 October 2026, released an open-weight version of its Muse Spark artificial-intelligence model, targeted at coding and AI agents, Reuters reported.
What Meta announced
- According to Meta, the model has 816 billion total parameters, of which 42 billion are active at any one time, and it can handle a context window of up to 1 million tokens.
- Meta says the model can run locally on two Apple M5 Ultra Mac Studio computers, and it is also offered through some API providers.
- In August, Meta had released Muse Glimmer, a smaller open-weight model.
Why “open-weight” matters
Open-weight models publish their trained parameters so developers can download, modify and adapt them, unlike proprietary systems that stay under their creators’ control. Reuters notes they can give businesses a cheaper alternative to closed systems and more control over their data — but they can also require significant computing resources and expertise to run at scale.
CEO Mark Zuckerberg has argued that advanced AI should not be concentrated among a small number of companies.
Performance claims above are Meta’s own; independent benchmarks were not part of the Reuters report.
Also available in: Саха тылаРусскийEspañol中文Portuguêsالعربية