Quick overview

---

Meta unveiled its AI model, the first since CEO Mark Zuckerberg started a multibillion-dollar overhaul of the company’s AI division to stay competitive. Under the direction of Chief AI Officer Alexandr Wang, Meta Superintelligence Labs, a new group of costly AI researchers, developed the eagerly awaited model known as Muse Spark.

Unlike the company’s prior open-source strategy, the Meta AI chatbot will be powered by Muse Spark, a closed model that means its design and code won’t be made public.  Meta stock rose by  6% in New York following the announcement.

The model is the first significant test for MSL, Zuckerberg’s new AI lab. The founder of Facebook hired Wang as part of a $14 billion investment in Scale AI last year after the company failed to keep up with rivals like OpenAI, Anthropic PBC, and Alphabet Inc. due to a string of setbacks. is Google

Meta has made an effort to maintain its AI division’s agility by granting researchers autonomy and reducing its usual management-heavy organizational structure. According to the executive, who wished to remain anonymous when discussing internal issues, Wang has about 100 direct reports. Although it was early in the company’s implementation, the executive admitted that Muse Spark lacked some of the capabilities of OpenAI’s ChatGPT, Anthropic’s Claude, or Google’s Gemini.

According to a blog post from Meta, the model is “an early data point on our trajectory,” with multiple larger models under development. Executives view Muse Spark—known internally as Avocado during development—as a revitalization of Meta’s AI strategy, which was previously centered on its open-source Llama models.

The project took nine months to complete. Although Meta still intends to develop open-source models in the future, Wang is an advocate of closed models, and the company is also thinking about offering API access to Muse Spark.