Meta Platforms has officially launched Muse Spark, a new artificial intelligence model designed to position the tech giant as a formidable competitor in the rapidly evolving AI landscape. The release marks a major milestone for the company’s artificial intelligence ambitions and represents its most significant update since the debut of its Llama 4 models over a year ago.
Developed by the newly formed Meta Superintelligence Labs, the Meta Muse Spark AI model is the first product to emerge from a costly restructuring effort. The launch follows a sweeping nine-month development cycle where the company rebuilt its entire artificial intelligence stack from the ground up, moving faster than any previous engineering phase in its history.
Massive Investments in the AI Race
The stakes for this launch are incredibly high, as major technology companies face growing pressure to demonstrate meaningful returns on their massive AI investments. Meta’s strategy involved a staggering $14.3 billion commitment in June 2025.
As part of this massive financial push, Meta brought in Scale AI chief executive officer Alexandr Wang to lead the Meta Superintelligence Labs. The company embarked on an aggressive hiring spree, recruiting top executives and engineers from industry rivals such as OpenAI, Google, and Anthropic. To secure top talent, Meta reportedly offered some engineers compensation packages worth hundreds of millions of dollars.
A Shift Away from Open-Source Development
One of the most notable aspects of the Meta Muse Spark AI model is a clear shift in the company’s overall strategy. Historically, Meta has been a champion of open-source technology, making the source code for its Llama models available to the public.
However, Muse Spark takes a different path. The technology is closed-source and is not available for public download. Instead, Meta has only shared a private preview of the system with unnamed partners. Currently, access to the new model is restricted exclusively to users in the United States.
Integration Across Social Media Platforms
The company is betting that applying advanced AI to everyday personal tasks will significantly boost engagement among the more than 3.5 billion users across its social media platforms.
Initially, the model is only available through the Meta AI application and website. In the coming weeks, Muse Spark will replace the existing Llama systems that currently power chatbots across the company’s ecosystem. Users will soon interact with the new model on WhatsApp, Instagram, Facebook, and Messenger. The technology will also make its debut on the Ray-Ban Meta smart glasses.
Meta has also teased new shopping features embedded directly within the chatbot. These tools point users straight to products they can purchase, offering a clearer picture of how the company intends to monetize its investments.
Practical Capabilities and Health Focus
Meta describes its new model as small and fast by design, while still being highly capable of reasoning through complex questions. The multimodal system can process both text and images to assist users with practical daily tasks. For example, a user can ask the system to estimate the calories in a meal from a photograph or superimpose an image of a coffee mug onto a physical shelf to see how it looks before buying.
The system is particularly focused on providing factual responses in domains like science, mathematics, and healthcare. To improve its health reasoning capabilities, the company collaborated with more than 1,000 physicians to carefully curate the training data.
Competing With Advanced Reasoning Modes
To handle the most complicated queries, the Meta Muse Spark AI model features a new “Contemplating mode.” This feature deploys multiple artificial intelligence agents to reason in parallel and work together simultaneously.
For instance, when planning a family vacation, one agent might draft the travel itinerary while another searches for kid-friendly activities. Meta states this mode allows its technology to directly compete with the extended reasoning capabilities of frontier models like Google’s Gemini Deep Think and OpenAI’s GPT Pro.
Performance and Future Outlook
Independent evaluations show that the new model is catching up with top systems from market leaders in language and visual understanding. However, tests compiled by the evaluation firm Artificial Analysis placed the model tied for fourth on a broad index, noting that it still lags behind competitors in areas like coding and abstract reasoning.
Meta leadership has been transparent about the system’s current limitations. Alexandr Wang acknowledged that there are rough edges in the model’s behavior that the team will polish over time. Larger and more advanced generations of the technology are already in development. Following the highly anticipated announcement, Meta shares saw a significant positive market reaction, climbing nearly 7 percent.
