Meta Unveils 'Muse Spark' — First AI Model from Its Superintelligence Lab
Meta unveils Muse Spark, the first AI model from its Superintelligence Lab led by Scale AI co-founder Alexandr Wang, featuring multimodal reasoning and multi-agent capabilities.
Meta has unveiled Muse Spark, the first AI model from its Meta Superintelligence Labs. The model represents the inaugural output from a new division created after CEO Mark Zuckerberg grew dissatisfied with Meta's AI progress. He recruited Scale AI co-founder and CEO Alexandr Wang, investing $14.3 billion for a 49% stake in Scale AI.
Muse Spark is a natively multimodal reasoning model supporting tool use, visual chain of thought, and multi-agent orchestration. A 'Contemplating' mode allowing multiple AI agents to collaborate on complex problems will be added later, enabling scaled test-time reasoning without drastically increasing latency.
The model will roll out to WhatsApp, Instagram, Facebook, Messenger, and Ray-Ban Meta AI glasses. Health-related question support is also planned, though the requirement to log in with a Meta account raises privacy concerns. Benchmarks show Muse Spark excels at visual STEM questions but lags rivals in coding ability. Meta's AI-related capital expenditures are projected to reach $115-135 billion in 2026, up from $72 billion in 2025.
Sources
Tools Mentioned in This Article
AI Newsletter
Get the latest AI tools and news delivered daily
Related Articles
OpenAI Launches GPT-5.4
OpenAI has launched its latest frontier model, GPT-5.4, designed for professional work. It introduces new features like real-time course correction during its thought process and supports up to 1M tokens of context. It outperforms experts in 44 professions and significantly reduces misinformation.
OpenAI Announces New Model for Professional Work, GPT-5.4
OpenAI has announced its latest AI model, GPT-5.4, specializing in professional work. It surpasses expert abilities in 83% of fields and reduces false claims by 62%, offering significant improvements in performance and reliability.
Perplexity Releases Open-Source Embedding Models to Compete with Google and Alibaba
Perplexity has released two open-source embedding models that rival Google and Alibaba, offering superior efficiency with less memory.