Meta introduces Muse Spark generative AI model

Meta has introduced Muse Spark, a new generative artificial intelligence model developed by Meta Superintelligence Labs (MSL). Designed to serve as a foundation for Meta AI, the model focuses on speed, reasoning, and multimodal capabilities, allowing it to process and understand various types of content, including text, voice, and images.

Muse Spark is being integrated across Meta’s ecosystem, including WhatsApp, Instagram, Messenger, Threads, and the company’s smart glasses. The technology utilizes multimodal pre-training, reinforcement learning, and test-time reasoning to handle complex tasks. It can also distribute complicated workloads among multiple AI agents working simultaneously.

For creators and businesses, the tool aims to simplify visual content creation through text commands, reducing the need for specialized technical skills or expensive software. Additionally, a version of the model supports programming tasks with a context window of up to 1 million tokens.