Meta The company, which is Facebook’s parent company, is a leading developer of OpenAI, Anthropic, google.
The AI model, named Muse Spark and originally codenamed Avocado, is the first in the company’s new Muse series announced Wednesday and was developed by Meta Superintelligence Labs, the AI division Wang oversees. Wang joined Meta in June as part of the company’s $14.3 billion investment in Scale AI, where he was CEO.
Meta is desperate to regain momentum in the competitive AI market following the disappointing debut of its latest open source model last April. The release failed to impress developers, prompting CEO Mark Zuckerberg to change strategy.
“Over the past nine months, Meta Superintelligence Lab has rebuilt our AI stack from the ground up, moving faster than any development cycle we’ve ever run,” Mehta said in a blog post on Wednesday. “This initial model is small and fast by design, but powerful enough to reason about complex questions in science, mathematics, and health. This is a strong foundation, and the next generation is already in development.”
Meta’s stock price soared nearly 9% on Wednesday, heading for its sharpest rise since January. Stocks rose and other markets followed suit as President Donald Trump announced a two-week halt to attacks on Iran and oil prices fell.
Meta doesn’t position Muse Spark as a top-of-the-line model, but emphasizes its efficiency and “competitive performance” in a variety of tasks.
Meta has been leveraging advances in generative AI and its own investments in the technology to strengthen its advertising business and improve efficiency across the company, but it has yet to crack the AI model market in a major way, and Space Zoom’s top competitors are booming. OpenAI and Anthropic are now worth more than $1 trillion combined, and Google’s Gemini technology and services are gaining traction, especially in the consumer market.
The stakes are huge, as the global generative AI market is estimated to grow more than 40% annually, rising from about $22 billion in 2025 to nearly $325 billion by 2033, according to Grand View Research.
Meanwhile, Meta is increasing spending on AI infrastructure to keep up with other hyperscalers. Meta said in its latest earnings report that AI-related capital spending in 2026 will be between $115 billion and $135 billion, nearly double last year’s capital spending.
The new Muse Spark is proprietary, and the company says it “hopes to open source future versions of this model.” The company took an open source approach to AI using the Llama family of models.
In a technical blog about the new model, Meta said that improved AI training techniques and a restructured technology infrastructure have enabled the company to create smaller AI models with capabilities comparable to older medium-sized Llama 4 variants with “orders of magnitude less computing.”
“Muse Spark provides competitive performance in multimodal perception, inference, health, and agent tasks,” Mehta said in the post. “We will continue to invest in areas where we currently have performance gaps, particularly in our long-term agent systems and coding workflows.”
New revenue opportunities
Meta is also experimenting with new AI model revenue streams by providing third-party developers with access to Muse Spark’s underlying technology via API. Currently, only unspecified “select partners” have access to a “private API preview” of the AI model, but Meta said it plans to eventually offer paid API access to a wider range of users at a later date.
The new model powers the company’s digital assistant with a standalone Meta AI app and desktop website. Muse Spark is set to debut within the coming weeks within Facebook, Instagram, WhatsApp, Messenger, and within the company’s Ray-Ban Meta AI glasses. Meta also plans for Muse Spark to eventually power the Meta AI app’s Vibes AI video capabilities. The service currently uses third-party AI models such as Black Forest Labs.
Alexander Wang speaks on CNBC’s “Squawk Box” outside the World Economic Forum in Davos, Switzerland, on January 23, 2025.
Jerry Miller | CNBC
Muse Spark allows users of standalone Meta AI apps and associated websites to switch between specific modes depending on the sophistication of their prompts. Users can tap one mode to get quick answers to simple questions, and tap another mode to perform more complex queries related to tasks like analyzing legal documents or gleaning nutritional information from photos of grocery store items.
Additionally, Contemplative Mode will be “rolled out gradually” in Meta AI apps and sites for the most complex queries and tasks, Meta said in a tech blog. In this mode, Muse Spark utilizes a team of AI agents to assist with “parallel inference,” helping it “compete with the extreme inference modes of frontier models such as Gemini Deep Think and GPT Pro,” the tech blog says.
The improved Meta AI using Muse Spark also includes a shopping mode, allowing people to buy clothes and decorate their rooms, the company says.
“Shopping mode draws from the styling inspiration and brand storytelling already happening across our apps, floating ideas from the creators and communities people already follow,” Mehta said.
WATCH: Alphabet, Meta, and Microsoft all fell due to increased data center spending.
