Meta Launches Muse Spark 1.1 with Enhanced Agentic AI Capabilities

Meta has announced the release of Muse Spark 1.1, a substantial enhancement to its multimodal reasoning model, now available for public preview through the new Meta Model API. This updated model is designed for agentic tasks, showcasing considerable improvements in its ability to utilize tools, interact with computers, generate code, and comprehend multimodal inputs. Muse Spark 1.1 demonstrates proficiency in complex, multi-application computer workflows, maintaining context across extended sessions and intelligently executing actions like scripting or direct UI interaction. It can navigate unfamiliar interfaces with minimal human intervention and orchestrate multi-agent systems, planning tasks across external applications and delegating work to subagents for faster project completion. Internally, Meta uses Muse Spark 1.1 for coding and research, where it performs competitively against leading models. Its capabilities extend to perception and multimodal reasoning, processing visual and audio inputs, and generating visual-to-code, rich captions, and agentic computer use, as demonstrated by its ability to create a Facebook Marketplace listing from a smartphone video.