Meta debuts first AI model from prized superintelligence group

Meta has launched Muse Spark, the first AI model developed by its newly formed Meta Superintelligence Lab (MSL), marking a key milestone in the social media giant’s aggressive AI push. The model, internally known as Avocado during development, was built over nine months and represents a strategic shift from Meta’s previous emphasis on open-source models like Llama toward proprietary, closed models designed to compete with leading AI developers such as OpenAI, Google, and Anthropic. This pivot is part of a broader investment and organizational overhaul championed by Mark Zuckerberg, who has made achieving artificial superintelligence a top priority for the company.

Muse Spark is viewed as a foundational model that will serve as the basis for future AI products and innovations across Meta’s platforms. The model’s performance is critical for establishing Meta’s AI credibility, especially as competitors continue to release increasingly advanced and powerful generative AI systems. Unlike prior Meta models that were openly shared with the developer community, Muse Spark’s closed-source nature signals Meta’s intent to monetize and tightly control its cutting-edge AI technologies.

The launch underscores Meta’s commitment to boosting its AI capabilities through heavy investment and reorganizing its talent under the MSL umbrella. This development puts Meta back into the competitive race in foundational AI research, aiming to challenge established leaders by delivering high-performance models that can drive next-generation AI applications and services at scale.