Gurugram, 9 Apr, (CXO Media): Meta Superintelligence Labs has launched Muse Spark, a multimodal reasoning model that will support its next phase of AI deployment across consumer platforms.
Muse Spark is the first model in the Muse family and is designed to handle text, images, and structured workflows within a single system. The model currently powers Meta’s AI app and website and is expected to be rolled out to platforms including WhatsApp, Instagram, Facebook, and Messenger in the coming weeks, indicating a broader push to embed advanced AI tools into widely used digital services.
The model introduces a feature called “Contemplating mode,” which allows multiple AI agents to operate simultaneously on complex tasks. This multi-agent design is intended to improve reasoning accuracy and efficiency, particularly for tasks that require layered problem-solving or visual interpretation. The system also supports tool use and visual reasoning, reflecting the growing shift toward multimodal AI systems capable of handling diverse data formats.
The company said Muse Spark was developed using a redesigned pretraining framework aimed at improving computational efficiency. More than 1,000 physicians contributed to refining the model’s health-related reasoning capabilities, and safety testing was conducted across key risk areas before deployment.
The launch reflects broader efforts by Meta Platforms to integrate advanced AI capabilities across its ecosystem, as technology firms increasingly focus on embedding reasoning-based AI systems into communication and digital productivity platforms.
