FOLLOW US

Meta Platforms (META.O) on Wednesday introduced Muse Spark, marking the debut of the artificial intelligence model created by a high-cost team assembled last year to compete in the AI arms race. Shares of the company rose nearly 7% following the announcement.

U.S. tech firms face mounting pressure to justify their massive AI investments. Meta’s stakes are especially high after hiring Scale AI CEO Alex Wang last year under a $14.3 billion deal, and offering some engineers compensation packages in the hundreds of millions to staff a new superintelligence team—a bid to regain prominence after its Llama 4 models fell short early last year.

Superintelligence refers to AI capable of outperforming humans. Muse Spark is the first of a series of models, internally called Avocado, produced by the new team.

The model, Meta’s first release in roughly a year, will initially appear only on the lightly used Meta AI app and website. In the weeks ahead, it will replace the Llama models powering chatbots on WhatsApp, Instagram, Facebook, and Meta’s smart glasses, the company said.

Meta did not reveal Muse Spark’s size, a standard measure for comparing AI computing power. Unlike prior open releases of Llama, Muse Spark was shared only as a “private preview” with select partners.

“This initial model is small and fast by design, yet capable enough to reason through complex questions in science, math, and health. It is a powerful foundation, and the next generation is already in development,” the company said in a blog post.

Independent evaluations found Muse Spark catching up to top models from Google, OpenAI, and Anthropic in areas like language and visual understanding, though it lagged in coding and abstract reasoning. The model tied for fourth place on a broad AI test index compiled by evaluation firm Artificial Analysis.

Rough Edges

Meta CEO Mark Zuckerberg had previously tempered expectations, telling investors in January that the team’s first models “will be good but, more importantly, will show the rapid trajectory that we're on.”

“I expect us to steadily push the frontier over the course of the year as we continue to release new models,” he said.

Wang, who leads the superintelligence team, acknowledged Wednesday on social media that “there are certainly rough edges we will polish over time in model behavior.” He added that larger versions are in development, with at least some planned for open release.

The launch also clarified Meta’s strategy for monetizing AI, teasing shopping features in the Meta AI chatbot that link users directly to products. Broadly, the company is betting that embedding AI into daily tasks will increase engagement among its 3.5 billion social media users, potentially giving it an edge over rivals with smaller audiences.

Muse Spark can assist with tasks such as estimating meal calories from a photo or superimposing a mug onto a shelf to preview its appearance, the company said. A Contemplating Mode, which runs multiple agents at once to enhance reasoning, allows Muse Spark to handle extended thinking tasks similar to Google’s Gemini Deep Think and OpenAI’s GPT Pro.

Meta said users could employ this mode for planning activities, like having one agent draft a travel itinerary while another searches for kid-friendly attractions.