Muse Spark 1.1 is a multimodal reasoning model from Meta positioned for agentic workflows that blend comprehension with action. Both external sources describe it as built for agentic tasks, combining a frontier reasoning architecture with broad input flexibility across text, images, MP4 video, MP3/WAV audio, and PDF documents. The model's reasoning mode is always on and cannot be disabled, with reasoning kept internal and billed as output tokens. Developers can tune the depth of this reasoning through a reasoning_effort setting that spans from minimal up to xhigh, allowing lighter calls for routine queries and heavier deliberation for complex multi-step work.

On the practical side, Muse Spark 1.1 returns text outputs and supports structured output, tool calling with auto or none tool_choice, and built-in web search that adds a small per-query cost and is reported in usage metadata. It accepts up to 50 attached images in one call, and PDFs contribute text from the first 100 pages plus page images from the first 50 pages. The very large context window makes the model well suited to long-document analysis, multi-attachment reasoning, and agentic pipelines where the model must keep substantial working memory across tool turns, rather than to narrow single-shot chat use cases.