The new unit is led by Alexandr Wang, whom Zuckerberg brought into the company through a $14 billion investment deal with Scale AI. Since then, Meta has poured billions more into recruiting top AI talent and committed tens of billions of dollars to infrastructure.

Today we’re introducing Muse Spark, our most powerful model yet, giving you a faster and smarter Meta AI.
Today we’re introducing Muse Spark, our most powerful model yet, giving you a faster and smarter Meta AI.

Still trailing rivals

Muse Spark will serve as a foundation for the Meta AI chatbot. Unlike Meta’s earlier open-source-oriented strategy, the model is closed-weight, marking a notable shift from the company’s previous Llama-based approach.

In some areas, Muse Spark still trails ChatGPT, Claude, and Gemini, though Meta says the project is still in its early stages. The company described the model as only the “first step on the journey,” with more powerful systems already in development.

Meta is positioning Muse Spark around efficiency and competitive performance across a range of tasks.

The model was built over a period of nine months and is being treated internally as a key part of Meta’s updated AI strategy. Previously, the company had focused heavily on the open-source Llama model family.

Muse Spark was trained using several third-party open-source tools, including Qwen from Chinese tech giant Alibaba, as well as technologies associated with OpenAI and Google.

“As with others in the industry, Meta uses distillation under strict safety measures,” a company spokesperson said.

Reasoning modes

Muse Spark includes several operating modes: Instant, Thinking, and Contemplating. The last of these is designed for deeper, research-grade responses. Meta says integration of Contemplating into the Meta AI app and web version will happen gradually.

The Contemplating mode uses a group of agents for what Meta describes as “parallel thinking.” Users can switch between modes depending on the complexity and nature of the prompt. Meta is also testing new monetization paths by offering third-party developers access to Muse Spark’s core technology through an API.

The company added that an updated version of Meta AI will include a dedicated Shopping mode to help users choose and purchase clothing.

Muse Spark is important less as a finished frontier winner and more as a signal that Meta is changing its AI playbook. The real story is the company’s move away from its open-model identity toward a more controlled, productized, and monetizable AI stack built around infrastructure, consumer features, and developer access.