Muse Spark ‘purpose-built’ for social media apps as investors question huge AI investment
Hannah Murphy and Cristina Criddle in San Francisco
Meta has launched its first AI model since chief executive Mark Zuckerberg led a multibillion-dollar spending spree on talent and infrastructure, in the first test of its efforts to catch up with rivals including OpenAI and Google.
The $1.6tn social media platform on Wednesday unveiled Muse Spark, which it said had been “purpose-built” for use across its products.
Zuckerberg signalled earlier this year that the release would not yet match the most advanced systems from rivals. But the company said Muse Spark would “power a smarter and faster” version of Meta AI, its virtual assistant.
It added that the model would enable more personalised and visual responses, drawing on content shared across Instagram, Facebook and Threads.
“Expect richer, more visual results, with Reels, photos and posts woven directly into your answers, with credit back to the content creators,” Meta said.
Meta’s shares were up 8 per cent on Wednesday following news of the model, amid a broad market rally.
It also said its new model had strong applications in healthcare. Meta said it had worked with more than 1,000 doctors to train the model to generate more detailed responses on topics such as nutrition and exercise.
Using the model, Meta AI will also offer a “shopping mode” to help users compare prices.
Unlike Meta’s previous Llama models, which have been “open” — or freely available for developers to tweak — Muse Spark is a smaller, closed model. The company intends to offer a “private preview . . . [to] select partners”. Meta said it hoped to open source future versions.
The release offers an early indication of how Zuckerberg plans to incorporate Meta’s social media content into its chatbot, as part of his promise to develop “personal superintelligence”.
It comes as the billionaire faces increasing pressure from investors to justify his colossal AI spending.
Over the past year, the chief executive has poured billions of dollars into building out a fleet of data centres and poaching rival AI researchers with packages reaching as much as 10 figures.
This is part of an all-out push to catch up in developing cutting-edge models after its previous model, Llama 4, fell below expectations last April, prompting Zuckerberg to carry out a rapid series of executive reshuffles and restructurings of Meta’s AI efforts.
The resulting “Meta Superintelligence Lab” is led by Alexandr Wang, 29, who joined after the Big Tech group invested $15bn in his data-labelling start-up Scale AI.
Within the new lab, Wang also leads a small, secretive team of elite AI researchers called “TBD” — To Be Determined — focused on spearheading the development of state of the art models.
Muse Spark beat leading models from Google, OpenAI and Anthropic in select benchmarks on reasoning and multimodal capabilities, according to evaluations shared by Meta.
But it acknowledged that the model was not yet cutting-edge in certain areas, adding: “We continue to invest in areas with current performance gaps, specifically long-horizon agentic systems and coding workflows.”
Zuckerberg said on Wednesday that Meta planned “to release increasingly advanced models that push the frontier of intelligence and capabilities, including new open-source models”.
“We are building products that don’t just answer your questions but act as agents that do things for you,” he added in a post on Threads.
With Muse Spark, Meta appears to lean into niche capabilities such as health, arguing that it is “one of the top reasons people turn to AI”.
Meta has also been working to improve the model’s “multimodal perception”, such as recognising objects in photos. Zuckerberg last year said Meta was doubling down on its investment in AI glasses, arguing wearable devices are vital to his bet on “superintelligence”.
“When Meta AI powered by Muse Spark comes to our AI glasses, the assistant will be able to better see and understand the world around you,” the company said.
