On Wednesday, Meta Platforms introduced Muse Spark, the inaugural artificial intelligence model developed by a high-profile team assembled last year as part of the company’s efforts to compete more effectively in the AI sector.

Following the announcement, Meta’s stock saw an increase, rising nearly 7% in value.

Large technology firms in the United States are facing increasing pressure to demonstrate that their significant investments in AI will yield substantial returns. This is particularly critical for Meta, which recruited Scale AI CEO Alex Wang last year through a $14.3 billion agreement and has offered lucrative compensation packages, amounting to hundreds of millions of dollars, to attract engineers for a new superintelligence initiative. This effort aims to re-establish Meta’s position among the leaders in AI following a lackluster performance with its Llama 4 models released earlier last year.

The term “superintelligence” describes AI systems that possess the capability to surpass human intelligence. Muse Spark represents the first release in a new series of models, internally referred to as Avocado, created by this specialized team.

This new model, which is Meta’s first launch in approximately a year, will initially be accessible exclusively through the underutilized Meta AI application and website. In the upcoming weeks, it will replace the existing Llama models that currently support chat functions on platforms such as WhatsApp, Instagram, Facebook, and Meta’s line of smart glasses.

Meta has not disclosed the size of Muse Spark, a crucial metric often used to evaluate the computing power of AI systems relative to competitors. Additionally, the company has shifted from its previous strategy of open releases for its Llama models, opting instead to provide only a “private preview” of Muse Spark to select, unnamed partners.

According to a blog post from the company, “This initial model is designed to be compact and efficient, yet sufficiently capable of addressing complex queries in areas such as science, mathematics, and health. It serves as a robust foundation for future developments, with the next iteration already in the works.”

Independent assessments of Muse Spark indicated that it is closing the gap with leading models from industry giants like Google, OpenAI, and Anthropic in certain domains, such as language processing and visual comprehension, while still trailing in areas like coding and abstract reasoning.

On a comprehensive index of AI performance tests compiled by the evaluation firm Artificial Analysis, Muse Spark ranked fourth.

Meta’s CEO, Mark Zuckerberg, had previously managed expectations regarding the model’s initial performance. In January, he expressed to investors that he anticipated the first models would be promising and would illustrate the rapid progress the team is making.

“I expect us to consistently push the boundaries throughout the year as we continue to unveil new models,” he stated.

Wang, who leads the superintelligence team, acknowledged in a series of social media updates on Wednesday that “there are certainly rough edges that we will refine over time in terms of model performance.” He also mentioned that larger versions of the model are under development and that Meta intends to release some of these versions publicly.

With this release, Meta has offered a clearer vision of how it plans to monetize its AI models, hinting at shopping functionalities integrated into its Meta AI chatbot that would direct users to products available for purchase.

Overall, the company is banking on the application of AI to enhance everyday tasks for its more than 3.5 billion social media users, which could provide a competitive advantage over rivals with a smaller user base.

Muse Spark is also capable of assisting users with various tasks, such as estimating the caloric content of a meal based on a photograph or overlaying an image of a mug on a shelf to visualize how it might appear in that setting.

An additional feature called Contemplating Mode, which allows multiple agents to operate simultaneously to enhance reasoning capabilities, would enable Muse Spark to engage in extended thought processes similar to those employed by Google’s Gemini Deep Think and OpenAI’s GPT Pro.

Meta indicated that this mode could assist users in efficiently planning family vacations, with one agent coordinating a travel itinerary while another searches for family-friendly activities.