Meta is taking another step in its AI strategy with the introduction of Muse Spark, the first model in Meta Superintelligence Labs’ Muse family.
With this model, the company aims to lay the foundation for what it calls a personal superintelligence. This is a form of artificial intelligence that doesn’t just answer questions, but actively thinks with you, understands you, and supports you in your daily life.
The Muse Spark comes at a critical moment for the meta. According to Reuters, this is the company’s first AI model in about a year and comes after expectations for the previous Llama 4 model were not met. Therefore, this announcement represents not only a technological advancement, but also an effort to reconnect with key populations in the AI market.
This model stands out because it was designed from the ground up as a multimodal inference model. Muse Spark can combine text, images, and other data types, and also supports the use of external tools and the simultaneous deployment of multiple AI agents. Meta therefore positions this system as the next step towards AI that can perform complex tasks independently and better understand context.
The introduction of Muse Spark also marks a broader overhaul of Meta’s AI approach. The company says it has been working in recent months on a new technology stack designed to help it scale more efficiently, from model architecture and training to infrastructure such as Hyperion data centers. Meta claims it can achieve comparable performance with significantly less computing power than previous models.
Reuters added that this technological leap requires significant investment. Last year, Meta assembled a new superintelligence team to keep up with its competitors, including through a multibillion-dollar investment in Scale AI and the appointment of CEO Alex Wang. The news agency said some engineers were lured in with exceptionally high pay packages, highlighting the company’s strategic priority towards AI.
In terms of performance, Meta claims Muse Spark is competitive across visual analytics, inference, and healthcare applications. At the same time, independent evaluations show that this model is still not at the top in all areas. External benchmarks show that Muse Spark performs well in language and visual understanding, but lags behind competitors in programming and complex abstract reasoning. In the overall ranking by the evaluation platform Artificial Analysis, this model is tied for fourth place.
A key new feature is a deliberation mode that allows multiple AI agents to work on a problem in parallel. This approach aims to enable deeper analysis without significantly increasing response time. Mehta sees this as a way to compete with the advanced inference modes of other major AI systems.
The applications Meta has in mind go beyond traditional chat interfaces. Muse Spark analyzes images, recognizes objects, and helps users with real-world tasks. This model also plays a role in the health field, such as extracting nutritional information from images or explaining bodily processes. To achieve this goal, Meta has worked with many doctors to improve the quality of answers.
At the same time, this announcement provides further insight into how Meta will monetize AI. The company is hinting at e-commerce integration that will allow users to view and purchase products directly within the chatbot. Additionally, Meta is focused on increasing engagement within its existing ecosystem, which currently includes billions of users.
Deployment of Muse Spark will occur in stages. Initially, the model will be available via the Meta AI app and website, and will then replace existing Llama models within platforms such as WhatsApp, Instagram, and Facebook in the coming weeks. Notably, Meta has not shared details about the scale of its model and is departing from its previous open strategy by offering only limited API previews to select partners.
