Meta releases Muse Spark 1.1 for better AI agent teamwork

Meta launched Muse Spark 1.1 on July 9, a multimodal reasoning model for AI agents. It improves planning, teamwork, and execution in agent tasks, plus better tool use, coding, and app control.

The new version boosts multi-agent collaboration: a main agent gathers info and plans, then splits tasks for sub-agents to run in parallel, cutting time on complex projects. It handles up to 1 million tokens of context, keeping key info over long workflows. For app operations, it runs long tasks across multiple apps, deciding whether to click interfaces directly, script automation, or do multiple steps at once. In coding, it diagnoses and fixes bugs, builds new features, and handles large-scale code migrations.

Meta says Muse Spark 1.1 passed safety checks under its internal framework, staying safe in chemical, biological, cybersecurity, and runaway risk areas, and better resisting prompt injection and jailbreak attacks. It's now live in Meta AI App and meta.ai's Thinking mode, with a preview API for developers to integrate into their apps.