Meta Unveils Muse Spark: First Model From SuperIntelligence Labs

Meta announced the release of Muse Spark on April 10, 2026, marking the first model produced by the company’s Meta Superintelligence Labs. The model is the debut entry in the Muse series, which Meta describes as a scientific approach to model scaling where each generation is used to validate and build upon the previous one before increasing the scale of the model.

The development of Muse Spark followed a nine-month period during which Meta Superintelligence Labs rebuilt its AI stack from the ground up. According to company documentation, this development cycle was faster than any previous cycle the organization had run.

Capabilities and Design

Muse Spark is designed to be small and fast, while maintaining the ability to reason through complex questions involving health, math, and science. Meta has characterized the model as a natively multimodal reasoning model, meaning It’s built to process and understand different types of data simultaneously.

A key feature of the model is its multimodal perception, which allows the Meta AI assistant to understand visual information rather than relying solely on text input. For example, a user can provide a photo of a snack shelf at an airport, and the AI can identify and rank items based on protein content.

Integration into Meta AI

On April 10, 2026, Meta updated the Meta AI app and the meta.ai website with a new visual design and the integration of Muse Spark. The assistant now allows users to switch between different modes depending on the complexity of the task at hand.

The updated assistant can now deploy multiple subagents in parallel to resolve a single query. In a travel planning scenario, for instance, one agent can draft an itinerary while a second agent compares different destinations, such as Orlando and the Florida Keys, and a third agent identifies activities suitable for children.

We are on our way to personal superintelligence: an assistant that can help anyone, anywhere with the things that matter most to them.

Meta Superintelligence Labs

Performance and Development

External analysis from The New York Times indicates that while Muse Spark performs better than Meta’s previous AI models, it currently lags behind its competitors in terms of coding ability.

Meta has stated that Muse Spark serves as a foundation for future iterations, and the next generation of the Muse series is already in development.