Meta Debuts Muse Spark, First AI Model from Its Superintelligence Team

Meta Platforms has introduced Muse Spark, the first artificial intelligence model developed by its new superintelligence team, marking a major step in the company’s effort to regain momentum in the fast-moving AI race.

The launch carries particular significance for Meta, which has invested heavily in rebuilding its AI standing after the underwhelming reception to its Llama 4 models. Last year, the company made a bold move by bringing in Scale AI CEO Alex Wang through a multibillion-dollar deal and offering highly competitive compensation packages to attract top engineering talent for its superintelligence unit.

Muse Spark is the first entry in a new internal model family known as Avocado. For now, the model will only be available through Meta AI’s app and website, but the company plans to roll it out across WhatsApp, Instagram, Facebook, and its smart glasses in the coming weeks, replacing the current Llama-powered chatbot systems.

Meta has not disclosed the model’s size, a detail often used to compare AI systems across the industry. In another notable shift, the company is also moving away from its earlier open-release strategy for Llama, offering only a private preview of Muse Spark to selected partners.

According to Meta, Muse Spark was designed to be lightweight and fast while still capable of handling complex reasoning tasks in areas such as science, mathematics, and health. The company described it as the foundation for future models, with more advanced versions already in development.

Early independent testing suggests Muse Spark is competitive with leading models from Google, OpenAI, and Anthropic in some areas, particularly language and visual understanding. However, it still trails behind in coding and more abstract reasoning tasks. Overall, it ranked tied for fourth on a broad benchmark compiled by evaluation firm Artificial Analysis.

Meta executives have acknowledged that the model is still evolving. CEO Mark Zuckerberg had previously told investors that the first releases from the team might not be groundbreaking immediately, but would demonstrate the company’s rapid pace of progress. Alex Wang also noted that the model still has “rough edges” that will be refined over time, while confirming that larger versions are in development and that some may eventually be released openly.

The launch also offers a clearer look at how Meta intends to monetize its AI efforts. The company previewed shopping features built into the Meta AI chatbot, allowing users to move directly from conversations to product discovery and purchasing.

More broadly, Meta is betting that AI-driven assistance for everyday personal tasks will deepen engagement across its massive global user base. Muse Spark is designed to help with practical requests such as estimating the calories in a meal from a photo or visualizing how an item like a mug might look on a shelf before purchase.

The model also includes a new “Contemplating Mode,” which uses multiple agents at once to improve reasoning. Meta says this feature can handle more advanced planning tasks, such as organizing a family vacation by having one agent draft an itinerary while another searches for child-friendly activities.