Meta Launches Muse Spark: Multimodal AI with Parallel Agents, Health Insights, and Shopping Intelligence

Meta has unveiled Muse Spark, its most powerful AI model to date, developed by Meta Superintelligence Labs. The model is built to handle complex reasoning across science, math, and health while introducing a range of practical consumer capabilities — from visual coding and shopping assistance to contextual search powered by real-time public social content.

One of Muse Spark's most distinctive features is its parallel subagent architecture. When given a complex task, the model can simultaneously launch multiple specialized agents working in concert. Asked to plan a trip, for example, Muse Spark might simultaneously draft an itinerary, compare destinations, and research activities — compressing hours of manual research into a single interaction. This multi-agent design positions Muse Spark as a step toward more autonomous AI workflows capable of handling open-ended, multi-step problems.

The model is natively multimodal, capable of analyzing images alongside text. A dedicated health assistance mode allows users to upload medical charts, lab results, or clinical images and receive detailed, physician-informed responses. Muse Spark can also generate functional websites and mini-games from text prompts, putting it in direct competition with OpenAI's GPT-5.4 in the creative coding arena. A shopping mode leverages Meta's vast social platform data to surface product recommendations and styling inspiration from creators and community trends — a capability no other major AI lab can match given Meta's unique social graph.

Muse Spark currently powers the Meta AI app and meta.ai website, with a rolling deployment underway to WhatsApp, Instagram, Facebook, Messenger, and Meta's AI glasses over the coming weeks. Meta described this as the first model in a new generation series, with each iteration validating the scientific advances of its predecessor before scaling further. Select API partners already have private preview access, and open-source versions are planned for future release. Meta Superintelligence Labs framed the launch as a deliberately scientific approach to model scaling — prioritizing depth of capability and reliability over raw parameter count.