Meta has finally unleashed the first AI model from its secretive Superintelligence Labs, the product of a multi‑billion‑dollar talent grab and infrastructure push as it races to close the gap with Google and OpenAI.

Muse Spark, described by Meta as its most powerful model yet, is already powering the Meta AI app and website, with the firm planning a wider rollout to WhatsApp, Instagram, Facebook, Messenger, and AI glasses in the coming weeks, as well as a private preview via API to select partners.

The tech firm said that the model’s development had taken place over the last nine months, with Muse Spark “small and fast by design”, and a “powerful foundation” for the next generation of the next wave of Meta AI models already in development.

Reeling off a list of capabilities, Meta said the AI will deliver personalised, visually rich responses by tapping into content across Instagram, Facebook and Threads, a vast reservoir of data unique to Meta’s platforms, allowing for a raft of features like a new “shopping mode”. 

The firm also emphasised Muse Spark’s potential in healthcare, saying that the Superintelligence team had worked with doctors to improve the model’s usability as a health tool, hoping perhaps to capture some of the 40 million daily users who turn to ChatGPT for healthcare queries.  

There was less said about the AI’s use as a coding tool, however, beyond Meta saying Muse Spark “excels” at tasks like creating custom websites or minigames, capabilities unlikely to challenge the likes of Claude Code or OpenAI’s suite of Codex tools.

Notably, during the firm’s most recent earnings call, Zuckerberg himself said that while he thought the first of Meta’s in-house AIs would be “good”, the company was more focused on pushing the boundaries of the technology over the coming year by merging it with the recommendation systems powering its social media empire.  

Recommended reading

• Oversight Board Slams Meta for Overlooking Fake AI Posts About Iran Conflict
• Will Meta Cut 20% of Its Workforce Over AI?
• Ex-Meta Engineer Accused of Mass Facebook Data Breach

Whatever those plans are, the firm will have to quickly step up the pace of its model development and deployment if it wants to keep up with rivals like Anthropic and OpenAI, as well as keep its investors on board with the mammoth spending that the Superintelligence Labs have so far demanded.

Meta’s CFO, Susan Li, told investors that the company’s 2026 capex would be in the range of $115-$135 billion, up from $72.22 billion in 2025, driven mainly by increased investments to support the Superintelligence Labs, which was also one of the main divisions behind Meta’s 6% jump in hiring over Q4’25.

Last year, Meta laid out about $14.3 billion for a 49% stake in the AI startup ScaleAI, with the deal seeing CEO Alexandr Wang joining Meta as the company’s Chief AI Officer, part of a talent poaching spree that also lured Apple’s head of Foundation Models, Ruoming Pang, in a $200m deal, as well as high-ranking OpenAI researchers, with their own hefty compensation packages.