Meta has shifted away from its open-source artificial intelligence strategy with the introduction of Muse Spark on April 8. This marks the company’s first fully proprietary AI model, moving decisively from the Llama approach that previously dominated its portfolio. The development comes from the newly formed Meta Superintelligence Labs, which was established following a $14.3 billion deal with Scale AI and led by Alexandr Wang.
Wang revealed that the AI stack underwent a complete rebuild over nine months, involving new infrastructure, architecture, and data pipelines. While he described Muse Spark as an initial step, he confirmed that larger models are in development and that future iterations may be open-sourced. However, the current model offers no public access to its weights, and API access is limited to selected partners via invitation only.
Industry analysts view this as a strategic pivot. Gartner’s Arun Chandrasekaran characterized it as a major shift, noting that Meta appears to be phasing out the Llama brand entirely. The change likely stems from competitive pressures, as both OpenAI and Anthropic profit significantly from their proprietary models—revenue that Meta could not capture with its open-source approach.
Muse Spark functions as a natively multimodal assistant, handling text, images, and voice. Its standout feature is a Contemplating mode that simultaneously runs multiple reasoning agents before delivering a response, putting it in direct competition with tools like Gemini Deep Think and GPT Pro. Meta also trained the model using data curated with over 1,000 physicians, positioning it as a personal health reasoning tool alongside general uses.
Performance-wise, Muse Spark trails behind GPT-5.4 and Gemini 3.1 Pro on the Artificial Analysis Intelligence Index, scoring 52 compared to their 57. However, the model outperformed Gemini 3.1 Pro on several health benchmarks that Meta emphasized during evaluation. The company has not disclosed the parameter count or detailed architecture.
The market reaction was immediate: Meta’s stock surged over 9% on launch day, marking its strongest single-day gain from a product announcement in more than two years. This comes as Meta plans capital expenditures of $115 to $135 billion for 2026, nearly double the previous year’s spending. Meanwhile, the developer community that built on Llama now faces uncertainty, as no confirmed timeline exists for the promised open-source release.