He explained that this model is optimized for agentic performance, coding, and utilization of tools. Muse Spark 1.1 is designed for extended tasks with a context window of 1 million tokens, capable of distributing tasks to sub-agents operating in parallel, and trained to navigate computer interfaces across desktop, mobile, and browser platforms.
(2) Muse Spark 1.1 excels in agentic performance, tool utilization, and computer interfacing. It effectively handles long-duration tasks with a 1M token context window, can allocate tasks to sub-agents in parallel, and is equipped to operate on desktop, mobile, or browser interfaces. pic.twitter.com/I3v82YohtR
— Mark Zuckerberg (@finkd) July 9, 2026
Zuckerberg further noted that the Meta Model API enables developers to leverage Muse Spark for the first time. Our aim is to provide robust agentic and multimodal models at minimal cost. More updates will follow soon.
According to Chief AI Officer Shengjia Wang, Meta’s Muse Spark 1.1, which was launched on Thursday, is the company’s most advanced model for agentic and coding tasks to date, as he stated in an interview with CNBC.
Wang also mentioned that the initial Muse Spark model, released in April, was accessible only to select partners via a private API preview.
Also Read: Meta to start AI chip production in September as it aims to double computing capacity: Report
