What Is Muse Spark? Meta's New AI Model Explained

Meta

unveiled Muse Spark on Wednesday, its new flagship artificial intelligence model and the first major product to emerge from the revamped AI division that Zuckerberg has poured enormous resources into over the past year. The model had been known internally as Avocado before its public debut. It is now live on Meta's standalone AI app, with a rollout to WhatsApp, Instagram and Meta's AI smart glasses coming in the weeks ahead.

How Good Is It?

On tests measuring writing and reasoning, Muse Spark came out significantly ahead of anything Meta has previously released. More notably, it landed close to, though not quite at, the level of the best models from Google, OpenAI and Anthropic, according to figures provided by Meta itself.The gap shows up most clearly in coding. That particular capability has become something of a benchmark in the broader AI competition, and Muse Spark still trails its top rivals there. It is a meaningful limitation in a race where coding ability is increasingly seen as a measure of a model's overall intelligence.

The Man Running It

Muse Spark is the first real test of Meta's new AI leadership. Zuckerberg brought in Alexandr Wang, a 29-year-old Silicon Valley entrepreneur, as the company's chief AI officer, part of a broader effort to replace the executives who oversaw Llama 4, Meta's previous flagship model that fell short of expectations when it launched a year ago. Several of those executives have since left the company, replaced by new hires drawn largely from OpenAI and other competitors.

A Closed Model, For Now

One notable shift with Muse Spark is how Meta is handling access to its underlying code. For years, Meta took pride in open sourcing its AI models, making the code publicly available for developers to build on. Muse Spark breaks from that tradition. The code is being kept private, at least for now. Meta said it may open source parts of the model down the line, but offered no firm commitment.

The Billions Behind It

The scale of Meta's AI ambitions is hard to overstate. Zuckerberg has committed to spending $600 billion on new data centers as part of his push to lead in AI development. This year alone, Meta expects to spend up to $135 billion, nearly double the $72 billion it spent last year, with AI sitting at the center of that investment.Last summer, Zuckerberg reframed the company's entire mission around building what he called "superintelligent" AI — technology capable of acting as the ultimate personal companion. To fund that push, Meta has laid off hundreds of workers and significantly scaled back its investment in the metaverse and virtual reality, areas that once defined the company's long-term ambitions.

Where Meta Stands in the Race

Anthropic, Google and OpenAI are still regarded as the leaders in foundational AI development. Muse Spark closes the gap but does not yet close it entirely. The announcement landed one day after Anthropic said its own latest model, Mythos, was considered too powerful to release safely due to cybersecurity concerns, a reminder that the competition at the frontier of AI is moving fast and carries real risks.For Zuckerberg, Muse Spark is both a product launch and a statement of intent. Whether it is enough to move Meta into the top tier of the AI race is a question the industry will be watching closely.