Meta Superintelligence Labs’ First AI Model Muse Spark Brings Reasoning, New Features to Chatbot

Meta Superintelligence Labs, a new division that handles the company's artificial intelligence (AI) research and development, released its first model on Wednesday. The release comes nearly nine months after the division was set up, marking an important moment in the Menlo Park-based tech giant's AI aspirations. Dubbed Muse Spark, the company claims it is faster and more capable than previous models. It is currently being rolled out to the Meta AI website and the app, with plans to integrate it into the company's social media platforms in the coming weeks.

Meta Introduces Muse Spark

In a newsroom post, the tech giant introduced Muse Spark and detailed its capabilities. The new AI model will primarily be used in Meta's products, but it will also be offered to select partners in private preview via an application programming interface (API). Currently, those interested can try out the model via the Meta AI app and the website.

Muse Spark is said to be a small model that prioritises output speed, but can also handle complex questions in science, mathematics, and health. Users can switch between Fast and Thinking modes depending on whether they need strong reasoning for the question. Additionally, the model lets the chatbot launch multiple subagents in parallel when the user sends complex questions that have multiple parts.

The new AI model also comes with improved vision capabilities. Meta claims users can upload any image, and the chatbot will be able to understand the context and respond with relevant information without the user having to explain the image. This feature will also be added to the company's AI glasses soon.

Additionally, Muse Spark is said to be able to answer questions about health more accurately. “We worked with a team of physicians to develop the model's ability to provide helpful information on common health questions and concerns,” the post said. Another capability with the AI model is visual coding. It can now generate custom websites and mini games based on text prompts.

Recommendation is another new capability added to the model. With this, Meta AI brings a shopping mode that helps users by surfacing styling inspirations and ideas for products. The products are curated from the creators and communities existing on the company's platforms.

The same feature can also be used to gain context during a conversation. The company says Meta AI can help explain the relevant context when users ask about a topic that is trending or a specific place. Additionally, users will also be able to tap the location to see public posts from locals who know the area. Similarly, by clicking the trending topic, users will see relevant content and community posts.