Meta unveils muse spark: Its new proprietary AI model built for personal superintelligence
- Marijan Hassan - Tech Journalist
- 1 hour ago
- 2 min read
3. AI
Meta unveils muse spark: Its new proprietary AI model built for personal superintelligence
Meta has officially announced the launch of Muse Spark, the first major release from its newly formed Meta Superintelligence Labs (MSL). Developed under the leadership of Chief AI Officer Alexandr Wang, Muse Spark represents a "ground-up overhaul" of Meta’s artificial intelligence stack, prioritizing speed, multi-modal reasoning, and seamless integration across its family of apps.
A Scientific Leap in Efficiency
Unlike previous iterations, Muse Spark was built using a new training pipeline focused on "thought compression." This reinforcement learning technique rewards the model for reaching accurate conclusions while penalizing excessive "thinking time."
The result is a model that Meta claims is 10 times more compute-efficient than previous versions, allowing for near-instant responses even when handling complex queries.
The model is currently available at meta.ai and within the Meta AI app, featuring two primary modes:
Instant Mode: Optimized for speed and casual day-to-day tasks.
Thinking Mode: Utilizes extended chain-of-thought reasoning for difficult math, science, and coding problems.
A third, more advanced "Contemplating Mode", capable of orchestrating multiple sub-agents in parallel, is expected to roll out in the coming weeks to compete with frontier models like GPT-5.4 and Gemini 3.1.
Natively Multi-modal
Muse Spark is natively multi-modal, meaning it was trained to understand text, images, and audio simultaneously. This architecture unlocks several practical, real-world features:
Visual STEM reasoning: The model can solve complex diagrams, troubleshoot home appliances via a photo, or identify nutritional information from an airport snack shelf.
Interactive creativity: Users can prompt the AI to generate custom websites, retro arcade games, or interactive dashboards that can be shared instantly with friends.
Advanced shopping: A new "Shopping Mode" leverages Meta’s vast product catalog and community styling inspiration to help users style rooms or find gifts based on current trends across Instagram and Threads.
Specialized focus on health
In a departure from general-purpose assistants, Meta has positioned health as a cornerstone of Muse Spark. Trained in collaboration with over 1,000 physicians, the model excels at "Health Reasoning." It can provide detailed nutritional mapping, such as visualizing diet-specific food recommendations with interactive annotations, and navigate complex health charts.
While Meta emphasizes that these tools are for informational guidance, the model significantly outperformed competitors on the HealthBench Hard evaluation.
Rolling out across the "meta-verse"
Meta plans to integrate Muse Spark into WhatsApp, Instagram, Messenger, and Facebook shortly. The model will also serve as the backbone for Meta’s smart glasses, where its visual perception capabilities are expected to be most impactful.
"Muse Spark is the first step on our scaling ladder," Meta stated in its official release. "We are on our way to personal superintelligence: an assistant that can help anyone, anywhere with the things that matter most to them."
