Meta launches Muse Spark

Meta just put a new AI model called Muse Spark inside its assistant, and the pitch is not “bigger chatbot.” The pitch is an assistant that can look at images, break a problem into parts, and send different sub-agents to work on those parts at the same time. (about.fb.com(about.fb.com)) To understand that, start with what a reasoning model is. A normal chatbot predicts the next word fast, while a reasoning model spends extra steps working through a problem the way a person might sketch on scrap paper before answering. (engadget.com(engadget.com)) Now add multimodal input. That means the system does not just read text; it can also inspect a photo, a chart, or a screen and connect what it sees to the question you asked. (about.fb.com(about.fb.com)) Then add tool use. Instead of answering from memory alone, the model can call outside tools the way a person might open a calculator, a browser tab, or a map app in the middle of solving something. (officechai.com(officechai.com)) Then add multi-agent orchestration. That is a fancy way of saying one model can spin up several smaller workers in parallel, with one worker comparing options, another gathering facts, and a third drafting the answer. (about.fb.com(about.fb.com)) Meta’s own example is family trip planning in Florida. One agent drafts the itinerary, another compares Orlando with the Florida Keys, and a third finds kid-friendly activities, then the system combines those results into one response. (about.fb.com(about.fb.com)) That is the backdrop for Muse Spark. Meta says the model is “small and fast by design,” but still able to reason through questions in science, math, and health, which tells you the company is aiming for speed plus capability instead of only chasing the largest possible model. (about.fb.com(about.fb.com); cnbc.com(cnbc.com)) Muse Spark is the first model in a new Muse series from Meta Superintelligence Labs. Meta says that team rebuilt its artificial intelligence stack over the last nine months, and CNBC reports the effort has been led by chief artificial intelligence officer Alexandr Wang after Meta’s June 2025 investment in Scale AI. (about.fb.com(about.fb.com); cnbc.com(cnbc.com)) The timing matters because Meta needed a reset. After the weak reception to Llama 4, the company was looking for a new path into a market where OpenAI, Anthropic, and Google had pulled ahead in both mindshare and product momentum. (cnbc.com(cnbc.com); theverge.com(theverge.com)) So Muse Spark is not just a lab demo. It already powers the Meta AI app and website as of April 8, 2026, and Meta says it will roll out to WhatsApp, Instagram, Facebook, Messenger, and its artificial intelligence glasses in the coming weeks. (about.fb.com(about.fb.com)) Meta is also opening a business lane. The company says it will offer Muse Spark in private preview through an application programming interface to select partners, which means outside companies will be able to plug the model into their own software without building a model from scratch. (about.fb.com(about.fb.com); cnbc.com(cnbc.com)) Meta’s product examples show the kind of jobs it wants this model to handle. The company says you can point the assistant at an airport snack shelf and ask it to rank the highest-protein options, or scan a product and ask how it compares with alternatives. (about.fb.com(about.fb.com)) The company is also tying the launch to a new safety process. Meta published an updated Advanced Artificial Intelligence Scaling Framework on April 8, 2026, and said Muse Spark went through a Safety and Preparedness evaluation before release, including checks across chemical, biological, cybersecurity, and loss-of-control risk categories. (about.fb.com(about.fb.com)) The bigger shift is strategic. For years Meta’s artificial intelligence identity centered on open Llama models, but Bloomberg reports Muse Spark is closed, and that suggests Meta now wants some of its most commercially important systems to live inside its own products and paid interfaces rather than only in the open model ecosystem. (bloomberg.com(bloomberg.com)) If Muse Spark works, Meta gets two things at once. It gets a smarter assistant across apps used by billions of people, and it gets an enterprise product that turns multimodal reasoning, tool use, and parallel sub-agents into infrastructure other companies can rent. (about.fb.com(about.fb.com); cnbc.com(cnbc.com))