Meta launches Muse Wonder and Muse Spark as closed-weight flagship AI models
Meta has introduced Muse Wonder and Muse Spark, a new family of flagship AI models named Muse, which succeed its Llama models. Unlike Llama, both Muse models are closed-weight and proprietary, so they cannot be downloaded or run locally.
Why it matters
Builders who relied on Meta's open-weight Llama models for self-hosting, fine-tuning, or local deployment will need to access Muse through Meta's hosted offerings or other channels instead. Meta's shift to closed models changes cost, data control, and vendor dependence considerations for AI-powered products.
Who should care
Developers and startups building AI features, teams that self-host or fine-tune Llama models, agencies choosing AI vendors for ecommerce and online business tools
What you can do
Audit which products depend on Llama weights, and test your key prompts against Muse Spark or Muse Wonder through Meta's access channels to compare quality, cost, and latency before migrating.