Meta introduced Muse Spark on April 8, 2026, marking the first release from Meta Superintelligence Labs (MSL), the company's new proprietary AI division led by Chief AI Officer Alexandr Wang. The multimodal reasoning model achieves performance equivalent to Llama 4 Maverick while requiring over an order of magnitude less compute for pretraining, representing a significant efficiency breakthrough in model development.
Muse Spark Demonstrates Advanced Multimodal Reasoning Capabilities
Muse Spark natively supports vision, tool-use, visual chain of thought, and multi-agent orchestration. The model features a "Contemplating mode" that orchestrates parallel agents for complex reasoning tasks, achieving 58% accuracy on Humanity's Last Exam and 38% on FrontierScience Research benchmarks. Through reinforcement learning techniques, the model demonstrates log-linear growth in pass@1 accuracy on training data, with gains generalizing to held-out evaluation sets.
Meta Superintelligence Labs Represents Strategic Shift in AI Development
Meta brought on Alexandr Wang as their first-ever Chief AI Officer in June 2025 after acquiring 49% of Scale AI for $14.3 billion. Wang, who founded Scale AI at age 19 and grew it to a $29 billion valuation, now leads MSL's development of proprietary models while Meta continues its open-source Llama releases. This dual-track strategy allows Meta to maintain competitive advantage with closed models while supporting the broader AI ecosystem.
Model Trained With Healthcare Expertise and Safety Guardrails
Muse Spark received training input from over 1,000 physicians specifically for interactive health information capabilities. Meta conducted extensive safety evaluations, finding the model "demonstrates strong refusal behavior across high-risk domains such as biological and chemical weapons" while maintaining safe margins across measured frontier risk categories.
Availability and Market Response
The model is available through meta.ai and the Meta AI app, with gradual rollout planned for WhatsApp, Instagram, Facebook, Messenger, and AI glasses. Select users gained access to a private API preview. Meta's stock rose 8.3% following the announcement, while the company's official announcement post received 6,815 likes, 788 retweets, and 1.36 million impressions.
Key Takeaways
- Muse Spark requires over an order of magnitude less compute than Llama 4 Maverick while achieving equivalent performance
- Meta Superintelligence Labs operates under Chief AI Officer Alexandr Wang, who joined after Meta's $14.3 billion acquisition of 49% of Scale AI
- The model achieves 58% accuracy on Humanity's Last Exam and 38% on FrontierScience Research benchmarks through parallel agent orchestration
- Over 1,000 physicians contributed training input for health information capabilities
- Meta stock rose 8.3% on the announcement as the company pursues a dual strategy of open-source Llama models and proprietary MSL releases