Last summer, Mark Zuckerberg handed Alexandr Wang—fresh from a $14.3B Scale AI acquisition—the reins to Meta Superintelligence Labs, complete with poached talent from top rivals and a blank-check budget. Nine months later, the first fruit: Muse Spark, a proprietary multimodal model handling voice, text, and images with a "contemplating mode" where AI agents debate tough problems.
Muse Spark shines in reasoning benchmarks, rivaling Opus 4.6 and GPT 5.4, and excels in health reasoning to fuel Meta's "personal superintelligence" vision. It trails in coding and ARC-AGI 2 but marks a leap from the open-source Llama family—Meta hints at future open-sourcing without a firm timeline.
This isn't just a model; it's Meta's no-holds-barred counterpunch in the AI race. After earlier LLMs flopped, Superintelligence rebuilt the stack from scratch, leveraging 3B+ daily users, vast data, and endless resources to claw back relevance.
Muse Spark offers the world agentic reasoning for real-world apps like health diagnostics and multimodal assistants, supercharging Meta's path to personalized AI woven into Facebook, Instagram, and WhatsApp. Big Tech without AI? Extinct.
META'S BACK—SPARKS FLYING, WAVES INCOMING!