Muse Spark: Meta's first AI model developed by the Superintelligence Lab

Meta’s new AI model, Muse Spark, is being positioned as the first visible step in a broader overhaul of the company’s artificial intelligence strategy, with the company saying it will expand capabilities over time and roll out new modes designed to handle more complex tasks.

From recruitment push to a new model

Press reports last July described Meta Superintelligence Labs (MSL) as a dedicated team within Meta focused on building general AI. The Wall Street Journal reported that the group is believed to include around 50 top researchers, with hires drawn not only from OpenAI but also from DeepMind, Anthropic, and Scale AI. The same report said Mark Zuckerberg personally took part in recruitment, inviting top talent to his home and joining other Meta executives in a recruitment event.

MSL was reportedly formed after Zuckerberg became dissatisfied with Meta’s AI progress, particularly as Meta’s Llama models have lagged behind OpenAI’s ChatGPT and Anthropic’s Claude. Meta has also named Alexandr Wang, co-founder and former CEO of Scale AI, to lead the effort.

What Meta says Muse Spark can do

In a press release last week, Meta said Muse Spark is now available on the web and in the Meta AI app, adding that the model is expected to improve over time. Meta also said it will introduce a “Contemplating mode” that enables the model to tackle more complex problems by using multiple AI agents in parallel, with the company saying this approach can reduce response time while improving reasoning without significantly increasing latency.

Meta is also incorporating health-related Q&A features, an area where several competitors are developing similar functionality. The company claims Muse Spark is especially effective for visually oriented STEM questions, including the ability to create interactive experiences such as small game-building and to help troubleshoot home devices.

Distribution across Meta products

Meta said Muse Spark is designed for its product ecosystem and is being integrated across platforms. The company said integration with Instagram, WhatsApp, Facebook, and Ray-Ban AI glasses is expected in the coming weeks. Meta also highlighted use cases such as shopping and travel planning, which it said are already common user needs on Instagram.

Competitive context and monetization questions

TechCrunch noted that some rivals place higher-capability models behind paywalls, but Meta has not clarified whether it will follow a similar strategy. The broader competitive backdrop includes OpenAI’s continued expansion of ChatGPT, Google’s expected launch of Android-powered smart glasses this year, and Apple’s plan to release an updated Siri with user-data-driven personalization.

Investment scale and ROI scrutiny

Coverage of Muse Spark has been broadly positive, with reporting pointing to the possibility that Meta’s large AI spending could begin translating into product value. However, investors are still assessing return on investment. In Meta’s January earnings call, Zuckerberg did not provide a specific ROI figure, saying the answer may not be satisfactory and describing the company as being in a phase of rebuilding its AI efforts.

Meta’s AI spending and related moves include a reported $14.3 billion investment in Scale AI in June of last year, along with the appointment of Alexandr Wang as AI chief. Meta has also acquired AI startups including Manus and Moltbook. OpenAI’s Sam Altman previously said Zuckerberg offered as much as $100 million to attract talent, and the article cited that Meta spent more than $72 billion on AI infrastructure in 2025.

Next steps

Looking ahead, Zuckerberg wrote on Threads that Meta expects to release increasingly advanced models in the coming months, including new open-source models. Meta also framed the effort as building products that go beyond answering questions—aiming to act as agents that can perform tasks for users.