Meta has unveiled its AI model Muse Spark, developed by the new team at Meta Superintelligence Labs (MSL).
Today we’re introducing Muse Spark, our most powerful model yet, giving you a faster and smarter Meta AI.
— Meta Newsroom (@MetaNewsroom) April 8, 2026
Muse Spark currently powers the Meta AI app and website and will be rolling out to @whatsapp, @Instagram, @facebook, @messenger, and AI glasses in the coming weeks.…
This is the first large language model (LLM) since CEO Mark Zuckerberg initiated a multibillion-dollar reorganization of the AI division to catch up with competitors.
The new department is led by Alexander Wang, who was brought on board as part of a $14 billion investment agreement with Scale AI. Since then, Meta has invested heavily in attracting specialized talent and developing infrastructure.
To date, the company has spent billions on recruiting AI talent and allocated tens of billions for infrastructure development.
Lagging Behind Competitors
Muse Spark will serve as the foundation for the Meta AI chatbot. Unlike previous products based on open-source technology, this model is closed.
In certain areas, the LLM falls short compared to ChatGPT, Claude, and Gemini, but the project is still in its early stages. The company emphasized that this model is just the "first step on the journey," with more powerful solutions in development.
Meta is focusing on the efficiency and "competitive characteristics" of Muse Spark for various tasks.
The neural network was developed over nine months and is viewed by leadership as a key component of a revamped AI strategy. Previously, the company concentrated on open-source Llama models.
Muse Spark was trained using several third-party open-source tools, including Qwen from Chinese tech giant Alibaba, as well as solutions from OpenAI and Google.
"Like other industry players, Meta employs distillation with strict safety measures," a company representative stated.
Thinking Modes
Muse Spark offers several operational levels: Instant, Thinking, and Contemplating. The latter provides scientific-level responses and will be integrated into the Meta AI app and web version gradually.
Contemplating utilizes a group of agents for "parallel thinking." Users can switch between modes based on the specifics of their requests.
Additionally, Meta is testing new monetization strategies by offering third-party developers access to the core technology of Muse Spark via API.
The company announced that a special "Shopping" mode will be added to the updated version of Meta AI, assisting users in selecting and purchasing clothing.
In February, Meta introduced the AI feature Dear Algo, allowing Threads users to personalize content recommendation algorithms.