Meta launches Muse Spark, its first model from new AI lab

Meta released an AI model on Wednesday called Muse Spark, marking the first step in a ground-up overhaul of the company’s artificial intelligence strategy. Muse Spark is the inaugural model to come out of Meta Superintelligence Labs. The creation of the new lab follows reports that Meta CEO Mark Zuckerberg was dissatisfied with the progress of the company’s previous models compared to competitors like OpenAI and Anthropic. To lead the new division, Meta recruited Alexandr Wang, the leader of Meta Superintelligence Labs.

To support this new division, Meta invested $14.3 billion in Scale AI for a 49% stake. Scale AI specializes in data labeling, which is the process of annotating data for AI training. This capital deployment highlights Meta’s push to secure data pipelines to compete with models from OpenAI, Anthropic, and Google. Zuckerberg emphasized the long-term vision for these developments on Threads, stating that the company plans to release increasingly advanced models to push capabilities, including new open-source models.

Muse Spark, which is now available on the web and the Meta AI app, is expected to improve over time. Meta plans to roll out a “Contemplating” mode designed to tackle more complex problems. To power this mode, the model uses multiple AI agents working simultaneously on the same problem. According to the company, “To spend more test-time reasoning without drastically increasing latency, we can scale the number of parallel agents that collaborate to solve hard problems,” Test-time reasoning is an AI processing method used to improve output quality. Zuckerberg noted that the ultimate goal is to build products that act as agents to perform tasks for users rather than just answering questions.

The company also indicated that Muse Spark could be applied to help users with health questions. However, this expansion brings regulatory and consumer trust challenges. Meta’s push into health — and even the more basic function of logging into Muse Spark — could raise privacy concerns. Users are required to log in using an existing Meta account, such as Facebook or Instagram, to access the model. While Meta has not explicitly stated whether it will train the model on personal data from these accounts, the integration of social media profiles with a health-capable AI assistant is likely to draw scrutiny.

Why it matters

Muse Spark represents a critical pivot for Meta, signaling a ground-up overhaul of its AI strategy driven by CEO Mark Zuckerberg’s dissatisfaction with previous progress relative to competitors like OpenAI and Anthropic.