Meta Launches Muse Spark: The First AI Model from Its Superintelligence Team
Meta Platforms introduced Muse Spark, marking the debut of the first artificial intelligence model developed by a high-profile team it formed last year to compete with rivals in the AI landscape. Shares of the company continued to rise, trading up nearly 7 per cent. US tech giants face mounting pressure to demonstrate that their substantial investments in AI will yield returns. The stakes are particularly elevated for Meta following its recruitment of Scale AI chief executive officer Alex Wang last year in a $14.3 billion agreement. The company has also extended compensation packages worth hundreds of millions of dollars to certain engineers to build a new superintelligence team, aiming to reclaim its position among the leaders in the AI sector after an underwhelming performance with its Llama 4 models earlier last year.
Superintelligence denotes AI systems that possess the capability to surpass human cognitive abilities. Muse Spark is the inaugural model in a new series, referred to internally as Avocado, from that team. The model, marking the company’s first release in approximately a year, will initially be accessible solely through the lightly utilized Meta AI app and website. The company announced that in the coming weeks, it will replace the existing Llama models that currently power chatbots on WhatsApp, Instagram, Facebook, and Meta’s collection of smart glasses. Meta has not revealed the size of Muse Spark, an important metric commonly utilized to assess an AI system’s computing capabilities against its competitors. It also altered its approach from earlier open releases of its Llama models, opting instead to provide only a “private preview” of Muse Spark to unnamed partners. “This initial model is small and fast by design, yet capable enough to reason through complex questions in science, math, and health.” The company said “It is a powerful foundation, and the next generation is already in development.”
Independent evaluations of Muse Spark’s performance revealed that it is catching up with leading models from market giants Google, OpenAI, and Anthropic in certain areas, such as language and visual understanding, while it still lags behind in others, including coding and abstract reasoning. The model achieved a tie for fourth place on a comprehensive index of AI tests assembled by the evaluation firm Artificial Analysis. Meta CEO Mark Zuckerberg had tempered expectations for early performance, telling investors in January that he believed the team’s first models “will be good but, more importantly, will show the rapid trajectory that we’re on.” He stated “I expect us to steadily push the frontier over the course of the year as we continue to release new models.” Wang, who leads the new superintelligence team, recognized in a series of social media posts on Wednesday that “there are certainly rough edges we will polish over time in model behavior.” He stated that larger iterations of the model were in the works and that Meta intended to release at least a portion of them publicly. With the release, Meta provided a clearer understanding of its strategy to monetize its models, hinting at shopping features integrated within its Meta AI chatbot that direct users to products available for purchase.
In general, the company is wagering that the integration of AI into daily personal tasks will enhance engagement among the over 3.5 billion users on its social media platforms, potentially providing it with a competitive advantage over rivals with a lesser reach. The company stated that Muse Spark can assist users with tasks like estimating the calories in a meal from a photo or superimposing an image of a mug on a shelf to visualize its appearance. An additional Contemplating Mode, which operates multiple agents concurrently to enhance reasoning capabilities, would enable Muse Spark to engage in the extended thinking modes of Google’s Gemini Deep Think and OpenAI’s GPT Pro. Meta stated that individuals could utilize the mode for effectively organizing a family vacation, with one agent creating a travel itinerary while the other researches kid-friendly activities.
