Meta’s Muse Spark 1.1 model excels at agentic tasks with 1M token context

Meta Superintelligence Labs has unveiled Muse Spark 1.1, a significant upgrade to its multimodal reasoning model designed specifically for agentic tasks. The model features a 1-million-token context window with intelligent compaction, allowing it to maintain coherence across extended sessions. In computer use, Muse Spark 1.1 autonomously decides when to write scripts for faster automation or click through interfaces directly, generating batches of actions at each step. It zero-shot generalizes to new native tools, MCP servers, and custom skills, and can orchestrate multi-agent systems as either a main agent (delegating to subagents) or a subagent (escalating when needed). This enables complex projects like organizing a dinner party while adapting to real-time changes without user intervention.

On the coding front, Muse Spark 1.1 shows substantial improvements in real-world tasks involving large codebases: diagnosing complex bugs, implementing new features in enterprise systems, and executing large code migrations. It integrates seamlessly with popular agentic coding setups, supporting planning mode, subagent delegation, and context compaction. The model also excels in multimodal perception and reasoning, generating visual-to-code artifacts and producing grounded outputs. Meta reports that internal developers use it daily to build faster, and researchers are automating model development and evaluation tasks. Muse Spark 1.1 is available now in 'Thinking' mode on the Meta AI app and meta.ai, while developers can access it through the public preview of the new Meta Model API.