Back to Headlines
Technology
Apr 09, 2026

Meta rolls out Muse Spark, the inaugural AI model from its $14.3 bn ‘superintelligence’ team, to challenge Google and OpenAI

AI Summary
Meta introduced Muse Spark, the first AI system produced by its high‑cost superintelligence unit led by former Scale AI CEO Alex Wang. The model, initially limited to Meta’s AI app, will replace Llama across its platforms and aims to compete with top models from Google, OpenAI and Anthropic, while also opening new revenue streams through integrated shopping features.

Meta announced the launch of Muse Spark, the debut artificial‑intelligence model from the company’s ambitious "superintelligence" squad that was assembled last year with a multi‑billion‑dollar budget.

The team, spearheaded by former Scale AI chief Alex Wang—brought on board in a $14.3 bn acquisition—has been offered compensation packages running into the hundreds of millions to attract top talent. Muse Spark is the first installment of the internally codenamed "Avocado" series.

For now, the model is accessible only through Meta’s AI app and website, but Meta says it will soon supplant the existing Llama models that power chatbots on WhatsApp, Instagram, Facebook and the firm’s smart‑glasses lineup.

Unlike earlier open releases of Llama, Meta has kept Muse Spark’s architecture details under wraps, offering a private preview to a select group of unnamed partners.

In a blog post, Meta described the system as "small and fast by design, yet capable enough to reason through complex questions in science, math and health," positioning it as a solid foundation for future, larger versions.

Independent testing shows Muse Spark narrowing the gap with leading models from Google, OpenAI and Anthropic in language and visual comprehension, though it still trails in coding and abstract reasoning tasks. The model placed tied for fourth on a comprehensive AI benchmark compiled by Artificial Analysis.

CEO Mark Zuckerberg had previously cautioned investors that early releases would be modest but would demonstrate a "rapid trajectory." Wang echoed this sentiment on social media, acknowledging "rough edges" that will be refined over time and confirming that bigger variants are already in development, with some slated for open release.

Beyond performance metrics, Meta hinted at commercial ambitions, embedding shopping suggestions directly into its AI chatbot to guide users toward purchasable items. With over 3.5 billion active users across its platforms, the company hopes AI‑driven personal tasks will boost engagement and create a competitive edge over rivals with smaller user bases.

Practical use‑cases highlighted include estimating meal calories from a photo, virtually placing a mug on a shelf via augmented reality, and a new "Contemplating Mode" that runs multiple agents simultaneously—mirroring advanced reasoning features seen in Google’s Gemini Deep Think and OpenAI’s GPT‑Pro.

Meta says this mode could, for example, help a family plan a vacation by having one agent draft an itinerary while another scouts kid‑friendly activities.