US tech giant Meta has introduced a new artificial intelligence model called “Muse Spark”, marking a strategic shift following a major internal overhaul of its AI division.

The model comes as a successor to Llama 4, released in April 2025 but widely seen as underperforming compared to global competitors. The setback prompted CEO Mark Zuckerberg to restructure the company’s AI teams.

With “Muse Spark”, Meta is betting on a smaller, faster model primarily designed for its own ecosystem. It is expected to be integrated across key platforms such as Facebook, Instagram, WhatsApp and Messenger, as well as its Meta AI assistant and smart glasses. For now, the model is only available in the United States.

Meta has also stepped up its investments in the AI space. In June, the company spent over $14 billion to acquire a 49% stake in Scale AI, a firm specializing in data preparation for artificial intelligence. Its CEO, Alexandr Wang, has since been appointed to lead Meta’s AI operations.

At the same time, Meta has recruited top talent from rivals including OpenAI, Anthropic and Google, bringing them into a newly created division known as the Meta Superintelligence Lab.

According to the company, “Muse Spark” is a multimodal model capable of processing both text and images. It can also handle complex queries in areas such as science, mathematics and health, while remaining more lightweight than competing systems.

Another key feature is its ability to generate computer code, enabling users to build simple websites or video games without advanced programming skills.

With this launch, Meta aims to regain momentum in the global AI race and lay the groundwork for more powerful models in the future.