Meta’s new Muse Spark model introduces multimodal perception and parallel sub-agents to WhatsApp, Instagram and Facebook applications.
Meta Superintelligence Labs has announced the launch of Muse Spark, the first in a new series of large language models designed to power the Meta AI assistant.
The model represents a rebuilding of Meta’s AI stack and is intended to support complex reasoning, mathematical tasks, and multimodal perception across the company’s suite of social applications.
Muse Spark is currently integrated into the Meta AI app and the Meta AI website, with plans to roll out to WhatsApp, Instagram, Facebook, Messenger, and Meta’s AI glasses in the coming weeks.
The model is described as a deliberate step in a scientific approach to scaling, in which each generation is intended to validate the last. While Muse Spark is designed to be small and fast, it is capable of handling complex questions in science and health.
A primary technical feature of the Muse Spark update is the introduction of parallel sub-agents. This allows the AI assistant to launch multiple processes simultaneously to address a single query. The model also features multimodal perception, enabling the assistant to understand visual information.
Meta has also collaborated with a team of physicians to enhance the model’s ability to navigate common health questions and provide detailed responses involving charts and images. Muse Spark also includes visual coding capabilities, allowing users to generate custom websites or mini-games through text prompts.
The company also expects to weave Reels, photos, and public posts directly into AI-generated answers, while providing credit to the original content creators. The roll-out will also include a strengthened risk framework and privacy protections as part of Meta’s broader goal of developing a personal superintelligence.
Stay updated on crypto and AI by following our socials
