Meta Muse Spark: First Proprietary Model from Meta AI Labs
3 min readMeta has officially unveiled Muse Spark, its first proprietary large language model developed under Meta Superintelligence Labs. The launch marks a decisive break from the company’s long-standing commitment to open-source AI and signals a new strategic direction for one of the world’s most influential tech companies.
From Open Source to Proprietary
For years, Meta built its AI reputation on Llama, a family of open-weight models that researchers and developers worldwide could freely download, fine-tune, and deploy. That approach made Meta a favorite in the open-source community and helped establish Llama as the backbone of countless AI applications. Muse Spark, by contrast, is closed and proprietary, available only through Meta’s own platforms.
The shift comes roughly nine months after Meta hired Alexandr Wang, founder of data labeling company Scale AI, as Chief AI Officer and tasked him with leading Meta Superintelligence Labs. Muse Spark is the first major model to emerge from that reorganized AI division.
What Muse Spark Can Do
Muse Spark is designed as a natively multimodal reasoning model. It handles text, images, and audio within a single system, and supports tool-use, visual chain-of-thought reasoning, and multi-agent orchestration. The model is intentionally compact and fast, optimized to run across Meta’s massive consumer ecosystem rather than as a research showcase.
The rollout started with the Meta AI app and website. Meta plans to bring Muse Spark to WhatsApp, Instagram, Facebook, Messenger, and its AI glasses in the weeks ahead. With billions of potential daily users across those platforms, the deployment scale is unprecedented for any single AI model.
On benchmarks, Muse Spark scores 52 on the Intelligence Index, placing it just behind Gemini 3.1 Pro and GPT-5.4, both at 57. Its standout area is health-related reasoning, where it scores 42.8 on HealthBench Hard, a result that aligns with Meta’s stated focus on consumer utility over raw capability comparisons.
Why This Shift Matters
Meta’s move toward proprietary AI is more than a product decision. It reflects a broader industry reckoning about whether open-source AI remains viable at the frontier. As training costs climb into the billions and competitive pressure intensifies, major labs are increasingly guarding their most capable models.
For users, the practical impact could be significant. A multimodal AI embedded across Meta’s entire platform suite would give billions of people access to advanced reasoning tools without requiring a separate subscription or app. That scale gives Meta leverage to shape how AI becomes part of everyday digital life, even if its model ranks second or third on leaderboard scores.
What to Watch Next
Meta says it plans to release open-source versions of Muse Spark eventually, though no timeline was given. Observers will watch closely to see whether the closed model improves quickly, how it performs on real-world tasks as it reaches mass deployment, and whether the open-source community responds by accelerating their own competing models.
As reported by CNBC and Axios, Muse Spark represents Meta’s most ambitious AI bet yet. Whether it pays off will depend less on benchmark scores and more on how well it integrates into the apps that billions of people use every day.