Meta Superintelligence Labs has released Muse Spark 1.1, a multimodal reasoning model built for agentic tasks such as coding, computer use, and multi-agent orchestration. The model supports a one-million-token context window and can act either as a main agent that plans and delegates work or as a subagent that executes a defined task and escalates when needed. Developers can access Muse Spark 1.1 through Meta’s newly launched Model API, now in public preview, as well as in “Thinking” mode within the Meta AI app. Meta said the model operates within its defined safety margins across frontier risk categories, including resistance to jailbreak attempts.
