Meta Unveils Muse Spark, First AI Model From New Superintelligence Team

Meta Platforms has introduced Muse Spark, its first artificial intelligence model developed by the company’s newly formed superintelligence team, marking a major step in its effort to regain momentum in the intensifying global AI race. The launch comes as major U.S. tech companies face growing pressure from investors to show returns on massive AI spending.

Meta Launches New AI Model After Llama 4 Setbacks

Meta announced Muse Spark on Wednesday as the first release from an expensive internal AI initiative assembled last year to help the company compete more directly with leading AI developers including Google, OpenAI, and Anthropic.

The company’s stock rose nearly 7% following the announcement.

Muse Spark is Meta’s first AI model release in roughly a year and follows disappointment surrounding its earlier Llama 4 models, which failed to meet expectations inside and outside the company.

Meta has invested heavily in the initiative, including a $14.3 billion deal involving Scale AI CEO Alex Wang, who now leads the superintelligence team. The company has also reportedly offered compensation packages worth hundreds of millions of dollars to attract top AI talent.

What Is Muse Spark?

Muse Spark is the first in a new internal family of AI models known at Meta as “Avocado.”

According to Meta, the model was intentionally designed to be smaller and faster than some rival systems while still capable of handling complex tasks in areas such as science, mathematics, and health.

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

The company did not disclose the model’s size, a standard metric often used to compare computing power across AI systems.

Limited Release and Shift Away From Open Models

Unlike its previous Llama releases, which were broadly open-sourced, Meta is taking a more restricted approach with Muse Spark.

The company said the model will initially be available only through the Meta AI app and website. In the coming weeks, it will replace the current Llama-powered systems behind Meta’s AI chatbots on WhatsApp, Instagram, Facebook, and the company’s smart glasses.

Meta also said it is offering only a “private preview” of Muse Spark to selected partners rather than publicly releasing the model.

That marks a notable shift in strategy for a company that previously championed open AI development.

Performance Trails Leaders in Some Key Areas

Independent benchmark testing suggests Muse Spark has narrowed the gap with top AI competitors in several categories, particularly language processing and visual understanding.

However, the model still lags behind leading systems in coding and abstract reasoning.

Evaluation firm Artificial Analysis ranked Muse Spark in a tie for fourth place on its broad AI benchmark index.

Meta Acknowledges “Rough Edges”

Meta CEO Mark Zuckerberg had already tempered expectations for the model’s early performance, telling investors in January that the company’s first releases from the team would be “good” but primarily demonstrate how quickly Meta intends to improve.

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

Alex Wang also acknowledged shortcomings in a series of social media posts Wednesday, writing that “there are certainly rough edges we will polish over time in model behavior.”

He added that larger versions of Muse Spark are in development and that Meta plans to release at least some future models more openly.

Meta Ties AI Push to Shopping and Everyday Tasks

The launch also offers a clearer look at how Meta plans to monetize its AI investments.

The company previewed shopping tools integrated into the Meta AI chatbot that can direct users to products available for purchase—part of a broader push to blend AI with online commerce.

Meta is betting that embedding AI into everyday digital experiences will increase engagement across its platforms, which collectively reach more than 3.5 billion users worldwide.

Muse Spark can also assist with practical tasks such as:

• Estimating calories in a meal from a photo
• Virtually placing household items, like mugs, on shelves using image overlays
• Helping plan travel itineraries and family vacations

New “Contemplating Mode” Targets Advanced Reasoning

Meta also introduced an enhanced reasoning feature called Contemplating Mode, which uses multiple AI agents simultaneously to improve performance on more complex tasks.

The feature is designed to compete with advanced reasoning systems such as Google’s Gemini Deep Think and OpenAI’s GPT Pro.

Meta said users could employ the mode for tasks like planning a family vacation, with one AI agent drafting an itinerary while another searches for child-friendly attractions.

Conclusion

Muse Spark represents Meta’s most significant AI launch in a year and a key milestone in its effort to reclaim a leadership position in artificial intelligence. While the model still trails top competitors in some advanced capabilities, its debut signals that Meta is accelerating development—and tying its AI strategy more closely to the products and platforms used by billions of people every day.

William Faulkner

“Social media evangelist. Baconaholic. Devoted reader. Twitter scholar. Avid coffee trailblazer.”