Meta just dropped its biggest AI product shift since Llama 4. The company announced Muse Spark on April 8, a new AI model built by Meta Superintelligence Labs that introduces selectable operational modes designed for different levels of cognitive demand.

The model represents a notable strategic pivot. Meta is moving away from its open-source AI playbook and toward hosted, integrated products embedded directly into its ecosystem of apps. The initial rollout is starting in the US across the Meta AI app, with expansion to Instagram, Facebook, WhatsApp, and Messenger, and Ray-Ban Meta smart glasses coming shortly after.

What the modes actually do

Muse Spark launches with two active modes and two more on the roadmap. The first, called Instant, does exactly what you’d expect: fast, lightweight responses for quick questions and simple tasks.

The second mode, Thinking, handles complex tasks like document analytics and nutritional assessments based on photos. Point your camera at a plate of food and the model will break down what it sees.

Two additional modes are planned for future release. Contemplating is designed for advanced multi-agent reasoning, suggesting Meta is building infrastructure where multiple AI agents collaborate on a single problem. Shopping, meanwhile, is geared toward e-commerce integration.

Some earlier reports referenced thematic labels like Deep Research, Presentations, and Social as planned additions to Meta AI. The official announcement, however, outlined a different taxonomy focused on operational capability rather than use case.

The strategic shift behind the model

Muse Spark is the first major release from Meta’s newly created Muse model family and the company’s first significant AI launch since Llama 4 arrived in 2025. Llama was open-source, designed to be downloaded and deployed by developers anywhere. Muse Spark is a hosted model tightly integrated into Meta’s own application ecosystem, with API access limited to a private preview for select partners at launch.

The model emphasizes what Meta is calling personal superintelligence, covering visual comprehension, health assistance, social media interactions, and online shopping. Leadership on the project includes Alexandr Wang among the key figures driving the initiative forward.

The competitive landscape and what investors should watch

OpenAI and Google have both been aggressively expanding their agentic AI capabilities. Meta is entering this race with direct access to the social graphs, messaging histories, and behavioral data of billions of users across its platform family.

For investors, the Shopping mode deserves particular attention. If Meta can successfully embed AI-driven commerce into WhatsApp and Instagram, it creates a direct monetization pathway that goes beyond advertising.

The Contemplating mode and Shopping mode are still listed as future expansions with no firm timeline, which means the initial product launches as a two-mode system.