Meta Releases Muse Spark, First AI Model from Dedicated Team

Meta CEO Mark Zuckerberg has announced the arrival of the company’s first artificial intelligence (AI) model, named Muse Spark. This model is the inaugural product from Meta Superintelligence Labs, a dedicated team established to develop AI, including efforts to catch up with competitors. In a post on his Facebook account, Zuckerberg stated that Muse Spark is now available and supports the latest version of Meta AI. Users can access it online through meta.ai or the Meta AI app. “Today, we are sharing our first achievement, Muse, our newest family of models. Spark is the first model in the Muse family, which you can try today,” wrote the man commonly known as Zuck, on Wednesday (8/4/2026) US time. Meta also describes Muse Spark as an initial step in their future AI development. Muse Spark is designed to assist users’ daily needs, such as visual understanding, health, shopping, and social media content. Moving forward, Meta aims to develop AI that not only answers questions but also acts as an agent performing various tasks for users. “Muse Spark is a built-in multimodal reasoning model with tool usage support, visual chain of thought, and multi-agent capabilities. It is the first step in our scalability ladder and the first product from our complete AI overhaul,” Meta stated on its official website. This concept was first revealed by Zuckerberg in July 2025, with the goal of delivering AI that helps individuals achieve their objectives, rather than being centrally controlled. As compiled by KompasTekno from Mashables, Meta is also preparing AI models in the Muse series family with open-source models accessible to the public. Meta has also released test results for Muse Spark on several popular benchmarks such as Humanity’s Last Exam (HLE), ARC AGI 2, and GPQA Diamond. The results vary; in several aspects, Muse Spark can surpass other AI models like Claude Opus, Gemini, GPT, and Grok. However, these results have not yet been independently verified. For certain conditions, Muse Spark excels in some aspects, such as question comprehension. Its score is quite high compared to several other models. However, in multimodal visual comprehension testing, its score still falls short compared to GPT 5.4.