Meta has launched Muse Spark, its first major AI model since the company overhauled its artificial intelligence division from scratch. The release, announced Wednesday, is the flagship product of Meta Superintelligence Labs, the unit led by Alexandr Wang, who joined the company in June as part of a $14.3 billion investment tied to his former company, Scale AI.
The launch sent Meta’s stock up nearly 9%, its sharpest single-day rally since January.
Muse Spark, which was internally codenamed Avocado during development, is the opening entry in a new product line Meta is calling the Muse series. The company said its team rebuilt the AI training pipeline from the ground up over a nine-month period, moving at a pace it described as faster than any previous development cycle.
What Muse Spark can actually do
Rather than positioning the model as a raw-power competitor, Meta is emphasizing speed, efficiency and practical usefulness. The model powers the company’s standalone Meta AI app and its desktop website, with a broader rollout to Facebook, Instagram, WhatsApp, Messenger and the Ray-Ban Meta smart glasses expected in the coming weeks.
Users can switch between different modes depending on what they need. A quick-response mode handles simple queries on the spot. A deeper reasoning option takes on more complex tasks, like parsing a legal document or reading nutritional information from a photo of grocery store products. A third feature, called Contemplating mode, deploys multiple AI agents working in parallel to tackle the most demanding requests, which Meta says makes it competitive with Gemini Deep Think and GPT Pro.
A shopping mode is also included, pulling styling ideas and product information from content already circulating across Meta’s platforms.
The model also supports multimodal tasks, meaning it can interpret images alongside text. According to Meta, if a user photographs a shelf of snacks at an airport shop, Muse Spark can identify the items and break down their nutritional details.
Meta’s AI spending reaches new heights
Muse Spark arrives at a moment when Meta is spending at a scale that would have seemed unthinkable just a few years ago. The company said its AI-related capital expenditures for 2026 will fall between $115 billion and $135 billion, nearly double the $72.2 billion it spent in 2025.
That investment reflects just how much pressure Meta is under to keep pace with rivals. OpenAI and Anthropic together carry a combined valuation of more than $1 trillion. Google’s Gemini products have made meaningful inroads with consumers. Meta’s previous open-source Llama models, released last April, failed to generate the excitement the company had hoped for, and the underwhelming reception forced a strategic reset.
Analysts expect Meta’s 2026 revenue to reach $247.7 billion, up from $198.8 billion in 2025, a 22% increase year over year.
A shift away from open source
Unlike the Llama family, Muse Spark will not be available as open-source software, at least not initially. Meta said it hopes to release open-source versions of future Muse models, but for now the technology will remain proprietary.
The company is also exploring a new revenue stream by offering third-party developers API access to Muse Spark’s underlying capabilities. A private preview is currently available to a limited group of partners, with paid access for a wider audience planned for a later date. The move signals a meaningful shift in how Meta thinks about monetizing its AI work.
