Meta Launches Muse Spark: The AI Model Built to Deliver Personal Superintelligence

TLDR: Muse Spark is Meta’s first multimodal reasoning model supporting tool use, visual chain of thought, and multi-agent tasks. Meta collaborated with over 1,000 physicians to strengthen Muse Spark’s health reasoning and medical response accuracy. Contemplating mode runs parallel AI agents, scoring 58% on Humanity’s Last Exam to rival top frontier AI models. Muse Spark uses ten times less compute than Llama 4 Maverick while delivering comparable performance across key benchmarks. Muse Spark, Meta’s newest AI model, marks a major step in the company’s push toward personal superintelligence. Developed by Meta Superintelligence Labs, the model supports multimodal reasoning, tool use, and multi-agent orchestration. It is now available at meta.ai and the Meta AI app. A private API preview is open to select partners. Meta also plans to open-source future versions of the model, widening access to its growing AI ecosystem. Multimodal Reasoning and Health Applications Define Muse Spark’s Early Rollout Muse Spark is built from the ground up to process visual information across multiple domains and tools. It performs well on visual STEM questions, entity recognition, and localization tasks. These abilities enable interactive experiences, from troubleshooting home appliances to building custom minigames. Meta positions this as a foundational part of its personal superintelligence roadmap. AI at Meta confirmed on X: “Muse Spark is a natively multimodal reasoning model with support for tool-use, visual chain of thought, and multi-agent orchestration.” Introducing Muse Spark, the first in the Muse family of models developed by Meta Superintelligence Labs. Muse Spark is a natively multimodal reasoning model with support for tool-use, visual chain of thought, and multi-agent orchestration. Muse Spark is available today at… pic.twitter.com/qnfSzoSPzt — AI at Meta (@AIatMeta) April 8, 2026 The model also introduces a health reasoning layer developed with input from over 1,000 phys...