Meta Platforms officially launched Muse Spark on April 8, marking a major shift in the company’s AI strategy. For the first time, Meta released a fully closed AI model, moving away from the open-source approach it used with Llama.
Muse Spark is the first product from Meta Superintelligence Labs, the new AI division built around Alexandr Wang after Meta’s massive $14.3 billion investment in Scale AI.
Wang said the team rebuilt Meta’s AI system from the ground up over the past nine months, including new infrastructure, architecture, and data systems. He also hinted that even larger models are already being developed.
Unlike Llama, Muse Spark’s model weights are not public. Access is currently limited to selected partners through invitation-only APIs. Meta says it may release open-source versions later, but for now the company is keeping the technology private.
The move surprised many developers because Llama became one of the most popular open-source AI models in the world, reaching around 1.2 billion downloads by early 2026. Analysts say this shows Meta now wants to compete more directly with companies like OpenAI and Anthropic, which make billions from closed AI systems and paid API access.
Muse Spark is designed to handle text, images, and voice at the same time. One of its main features is a new “Contemplating” mode that runs several reasoning processes together before giving an answer. Meta is also promoting the model as a health-focused AI assistant after working with more than 1,000 doctors to help build its medical training data.
The company says Muse Spark will soon be integrated across WhatsApp, Instagram, Facebook, and Messenger. Investors reacted positively to the announcement, with Meta shares jumping more than 9% on launch day.
Even though Meta says future versions could become open source again, many developers who built projects around Llama are now waiting for answers about the company’s long-term plans.
