Meta on Wednesday launched muse spark, describing it as its most powerful AI model yet and a central piece of its wider AI rebuilding effort. the model is the first release from its Meta Superintelligence Lab, the unit Mark Zuckerberg unveiled in June in response to criticism that earlier AI systems fell short. Meta is already using muse spark inside Meta AI services, and executives say larger models are in development.
Meta puts muse spark at the front of its turnaround
The launch places muse spark at the heart of Meta’s latest effort to reset its standing in artificial intelligence. Zuckerberg created the Meta Superintelligence Lab after the company faced criticism over the performance of prior models, and he later brought in several AI leaders, including Alexandr Wang, the founder of Scale. AI.
Wang said on X that Meta is already using muse spark to power Meta AI services and is working on larger models. In a separate Facebook post, Zuckerberg said the system is “a world-class assistant” and highlighted its strength in visual understanding, health, social content, shopping, games, and related personal superintelligence tasks.
Meta said muse spark can assign multiple subagents to different tasks at once, allowing it to handle more complex requests in parallel. In one example, users will be able to ask Meta AI to help plan family trips by drafting an itinerary, comparing destinations, and identifying possible activities at the same time.
What Meta says the model can do
Meta also said muse spark can recognize images and draw insights from what it sees. The company gave an example of a user taking a picture of snacks on a shelf in an airport shop, after which Meta AI could identify the items and point out which has the most protein or calories.
Wang added that muse spark’s contemplating mode is competitive with other extreme reasoning models such as Gemini Deep Think and GPT Pro. In an evaluation document, Meta said the model tops AI systems from Anthropic, Google, OpenAI, and xAI in some categories, while trailing in others.
The company is not making muse spark open source, unlike some earlier AI models, though Meta said it hopes future versions will be open. That choice marks a notable shift as Meta tries to balance broader AI access with tighter control over its newest systems.
Investors watch the spending story
Meta’s AI push comes with heavy spending. During its most recent quarter, it expects to spend between $115 billion and $135 billion in 2026, up from $72. 22 billion in 2025.
Still, the financial backdrop is giving the company some room. Meta’s 2025 revenue reached $198. 8 billion, up 22% from $162. 4 billion the year before, and analysts expect 2026 revenue to climb to $247. 7 billion. Meta’s stock rose more than 9% in midday trading after the muse spark launch, signaling that investors are watching the turnaround closely.
The next test will be whether muse spark can translate early technical claims into broader product gains inside Meta AI services. If the company can keep improving the model while scaling larger versions, muse spark could become a key marker of how far Meta’s AI turnaround can go.
