Over the past year, Mark Zuckerberg has devoted his soul, body, and wallet to relaunching Meta’s artificial intelligence, which had fallen behind competitors. The company has allocated between $115 billion and $135 billion for 2026, with infrastructure costs as the main expense, in addition to recruiting the crème de la crème of experts in the field, starting with the former co-founder and CEO of Scale AI, Alexandr Wang, to lead the Superintelligence Labs. Now, after nine months, Muse Spark, their first AI model, has been released and the moment of truth has arrived.
Although it is smarter and faster than previous models in the Llama series, Muse Spark is not yet comparable to competitors’ models in programming, that is, code generation, which is the area where all companies in the sector are focusing most of their resources.
Furthermore, since users will need to log in with an existing Meta account, such as Facebook or Instagram, to use it, privacy concerns arise. Meta does not explicitly state that personal information from these accounts will be used by the AI, but according to TechCrunch, it is likely, considering that the company generally trains its models on users’ public data and has presented Muse Spark as a personal superintelligence product.
FEATURES AND FUNCTIONALITIES OF MUSE SPARK
Muse Spark is the first model of the new Muse family, designed for speed and efficiency rather than size, and, as explained by Quartz, represents a strategic shift compared to previous open-source Llama models. Muse Spark powers the Meta AI chatbot on the app and meta.ai website and will soon be integrated into Facebook, Instagram, WhatsApp, Messenger, and the Meta AI Ray-Ban smart glasses.
It accepts voice, text, and image inputs but produces only text output, writes Axios. It offers three reasoning modes: Instant, Thinking, and Contemplating, with the latter allowing multiple AI agents to work in parallel on complex queries and, according to Mashable, “enables Muse Spark to compete with the extreme reasoning modes of leading models like Gemini Deep Think and GPT Pro.”
The model also includes a Shopping mode that combines data on user behaviors and interests with content from creators and brands to provide personalized recommendations on Meta platforms. Additionally, according to Zuckerberg’s company, Muse Spark demonstrates excellent reasoning capabilities in the health domain, thanks to collaboration with over 1,000 doctors to improve reliable and safe responses.
HOW IT STACKS UP AGAINST THE COMPETITION
Muse Spark, reports Bloomberg, does not yet reach the levels of competing models in coding, a sector where OpenAI, Anthropic, and Google are focusing the most resources, but excels in science, health, and mathematics, as well as multimodal and agentic tasks. OpenAI and Anthropic, notes Cnbc, are collectively valued at over $1 trillion, while the global generative AI market is expected to grow more than 40% annually, rising from about $22 billion in 2025 to nearly $325 billion by 2033.
Meta has stated that Muse Spark “offers competitive performance in multimodal perception, reasoning, health, and agentic tasks” and will continue to invest in long-range agentic systems and programming workflows.
THE (USUAL) PRIVACY CONCERNS
Access to Muse Spark requires an existing Meta account, such as Facebook or Instagram, which immediately raises concerns about privacy and control over personal data. Meta does not explicitly clarify whether information from users’ accounts will be used to train or personalize the AI, but according to TechCrunch, it is very likely, considering the company has historically made extensive use of users’ public data for training its models. This choice therefore raises questions about the protection of sensitive data and possible user profiling, especially in the health domain, where Muse Spark provides responses based on personal and medical information.
Although the model is free, Meta is considering introducing paid subscriptions and API access for selected partners, which could create disparities in access and increase centralization of control over generated data. This combination of mandatory access via proprietary accounts and potential future limitations suggests that Muse Spark, while innovative, carries significant risks related to privacy management and governance of user data.
