Meta launches Muse Spark 1.1 AI model with expanded agentic capabilities Proactive uses images sourced from Shutterstock
Meta Platforms Inc (NASDAQ:META, XETRA:FB2A, SIX:FB)‘s Meta Superintelligence Labs on Thursday announced the launch of Muse Spark 1.1, an updated multimodal reasoning model designed for advanced artificial intelligence workflows, including autonomous agent tasks, coding and computer-use applications.
The release expands Meta’s push into the enterprise AI market, with the company also opening a public preview of the Meta Model API, allowing external developers to access Muse Spark 1.1 for the first time.
Meta said Muse Spark 1.1 features a one-million-token context window and improvements in tool use, multimodal reasoning and coding performance.
The company positioned the model as capable of handling complex workflows by operating within multi-agent systems, where it can act as both a primary agent coordinating tasks and a subagent completing specialized assignments.
The model is currently available in “Thinking” mode through the Meta AI app and meta.ai, while developers can access it through the Meta Model API.
According to Meta, Muse Spark 1.1 is designed to manage long-running tasks by retaining relevant context, retrieving information from earlier parts of a workflow and coordinating actions across multiple tools. The company said the model can navigate computer interfaces, automate tasks, write scripts when appropriate and adapt to changing requirements during extended sessions.
Meta also highlighted improvements in software development capabilities, saying Muse Spark 1.1 can assist with debugging, feature development and large-scale code migrations. Internal testing on the company’s Meta Internal Coding Bench showed improvements over the previous Muse Spark model and competitive performance against other leading AI systems, Meta said.
The company added that Muse Spark 1.1 offers expanded multimodal capabilities, including the ability to interpret images, video and other visual information while using those inputs in agent-driven tasks.
Meta said the model underwent safety evaluations focused on areas including cybersecurity, chemical and biological risks, and potential loss-of-control scenarios before deployment. The company reported that the model remained within its safety thresholds across those evaluations.
Shares of Meta were little changed at $605 following the update, having fallen more than 8% so far this year.