Muse Spark, the first large language model from Meta Superintelligence Labs, launched on April 9, 2026, bringing multimodal reasoning and parallel agent capabilities to Meta’s family of apps and services.
Meta built the model over nine months, rebuilding its AI infrastructure from the ground up. The company said this was the fastest development cycle it had completed. Muse Spark is the first in a new series called Muse, where each generation is validated before the next, larger model enters development.
What Muse Spark Brings to Meta AI
Muse Spark currently powers the Meta AI app and the meta.ai website. The model supports two operating modes: Instant, for quick responses, and Thinking, for complex reasoning tasks. Meta AI can now run multiple subagents in parallel, so a single query, such as planning a family trip, can simultaneously produce an itinerary, a destination comparison, and a list of activities.
The model will roll out to WhatsApp, Instagram, Facebook, Messenger, and Meta’s AI glasses in the coming weeks. A private preview via API is also available to select partners, and Meta said it hopes to open-source future versions of the model.
Multimodal Perception and Health Features
Muse Spark includes strong multimodal perception, allowing Meta AI to analyze images rather than relying solely on typed input. A user can photograph a product shelf and receive a ranked comparison, or scan an item to see how it measures against alternatives. Meta said these perception capabilities become more useful when the model runs on AI glasses, where the assistant can observe the user’s surroundings directly.
Health is a specific focus area. Meta worked with a team of physicians to develop the model’s ability to respond to common health questions, including queries that involve images and charts. The company said health is one of the top reasons people turn to AI assistants.
Shopping Mode and Social Context
Meta AI now includes a shopping mode that draws from styling content and brand posts already circulating across Instagram, Facebook, and Threads. The feature surfaces product ideas based on creators and communities a user already follows. Shopping mode is launching first in the United States, with expansion to additional markets planned for later.
When users search for a location or trending topic, Meta AI pulls relevant public posts from locals and community members into the conversation. This gives users real-time context from people connected to the subject, directly within the social media experience they already use.
Visual Coding and Rollout Timeline
Muse Spark also supports visual coding, letting users generate custom websites and mini-games from text prompts through Meta AI. Examples include building a party-planning dashboard, a retro arcade game, or a flight simulator, all shareable with other users.
The upgraded Meta AI experience, including both Instant and Thinking modes, is rolling out in the United States first. Meta said it plans to expand to more countries and platforms in the coming weeks. Future updates will integrate Reels, photos, and posts directly into AI responses, with attribution to the original content creators. The company also said it is strengthening its safety and privacy frameworks alongside the model’s release.
“We are building toward personal superintelligence, an AI that does not just answer your questions but truly understands your world because it is built on it.”
Meta Superintelligence Labs
