Superintelligence Labs Releases Muse Spark 1.1, a Multimodal Agent Model
Superintelligence Labs has released Muse Spark 1.1, a multimodal model designed for code generation, agentic workflows, and computer control. Unlike the Llama family, the model’s weights remain closed. Acting as a chief agent, Muse Spark 1.1 gathers context, builds a plan, and delegates tasks to sub-agents, all within a 1 million token context window—the model can also compress past session history and integrate with MCP servers. In the MCP Atlas and Humanity’s Last Exam benchmarks, Muse Spark 1.1 surpassed Opus 4.8, GPT 5.5, and Gemini 3.1 Pro, with Opus retaining the lead only in SWE-Bench Pro.
Alongside the model, an API has been launched at $1.25 per million input tokens, $4.25 per million output tokens, and $0.15 per million cached input tokens. The model is available in Thinking mode via the web interface and Meta AI app.