Meta announced on Thursday the next version of its multimodal Spark AI model, Muse Spark 1.1, built specifically for agentic tasks with promised gains in computer use, coding and multimodal understanding. The model was released by Meta’s Superintelligence Labs, led by former Scale AI CEO Alexandr Wang. Meta also launched a public preview of its Meta Model API, making Spark 1.1 accessible to developers, and it is now available in the "Thinking" mode within the Meta AI app and on meta.ai. Meta says Spark 1.1 can complete complex projects significantly faster than its predecessor and is trained to "orchestrate multiagent systems to optimize end-to-end latency." This multiagent system features a main agent that makes a plan and delegates execution across parallel subagents. Spark 1.1 actively manages its 1-million-token context window by remembering actions, retrieving information and compacting context. The model may also take control of a computer for certain tasks, maintaining context across long sessions and navigating unfamiliar interfaces. On coding, it promises improvements in tasks involving complex databases, including diagnosing bugs, adding features and performing large code migrations. Its multimodal capabilities allow it to ingest imagery and audio to provide "ultra-descriptive" captions.