Details
- Meta Superintelligence Labs introduced Muse Spark 1.1 in July 2026, a major upgrade to Muse Spark and a multimodal reasoning model optimized for agentic tasks, tool use, computer use, and coding.
- The model features a 1 million-token actively managed context window, native multi-agent orchestration, and strong performance on complex coding, multimodal perception, and real-world computer-use workflows.
- Muse Spark 1.1 is available in "Thinking" mode in the Meta AI app and on meta.ai, and for the first time via the new Meta Model API public preview, which offers OpenAI-compatible endpoints and paid access with free trial credits.
- Compared with the original Muse Spark released in April 2026, 1.1 delivers substantially faster project execution, better context management, and higher scores on agentic benchmarks such as MCP Atlas and JobBench, positioning it against frontier models from OpenAI and Anthropic.
- Early partners including Replit, Cline, and Box highlight Muse Spark 1.1’s combination of long context, multimodal inputs (images, video, PDFs), strong coding and reasoning, structured output, and parallel tool calling as a complete foundation for large-scale agentic applications.
Impact
Muse Spark 1.1 and the Meta Model API move Meta from primarily consumer-facing assistants into direct competition for paid developer workloads in agentic coding and automation. By combining long-context multimodal reasoning with aggressive pricing and OpenAI-compatible tooling, Meta is likely to accelerate adoption of agentic architectures, push competitors to improve tool-use and orchestration capabilities, and influence R&D roadmaps toward large-context, workflow-native models over the next 12–24 months.
