Muse Spark 1.1 is a multimodal reasoning model in Meta's muse family that reads text, images, PDFs, and video and returns text, letting a single system handle screen captures, diagrams, scanned documents, and UI frames alongside ordinary code and prose. It ships as a closed-weights service exposed through LLM Gateway, with reasoning, tool calling, structured output, attachment handling, and temperature control built in, so it can plan tasks, delegate to subagents, write scripts, click through UIs, and debug using screenshots through an OpenAI-compatible endpoint. Practical fit centers on long, mixed-media agentic pipelines that need a very large working memory and reliable tool orchestration, rather than purely conversational chat.
In comparative coding evaluations published after its July 2026 update, Muse Spark 1.1 stands out for agentic tool use on JobBench and MCP Atlas while trailing on pure coding benchmarks such as SWE-Bench Pro, where GPT-5.6 Sol leads on Terminal-Bench and accuracy. Its pricing positions it aggressively against Claude Fable 5, undercutting that rival on input cost while keeping a million-token context window that supports extended debugging sessions. The model fits teams building code agents, document-aware assistants, and multimodal automation that need strong tool orchestration and long-context recall more than top raw code-completion scores.
