Meta Launches Muse Spark 1.1, A Multimodal Reasoning Model Built For Agentic Tasks

By Amit Chowdhry • Today at 1:01 PM

Meta Superintelligence Labs has released Muse Spark 1.1, a significant upgrade to Muse Spark and the latest model from the research division Meta created to pursue its personal superintelligence vision. Muse Spark 1.1 is a multimodal reasoning model built for agentic tasks, with major performance gains in tool use, computer use, coding, and multimodal understanding. The model is now available in “Thinking” mode in the Meta AI app and on meta.ai, and is accessible to developers via the Meta Model API, currently in public preview.

Muse Spark 1.1 is designed to plan and orchestrate actions across external apps and services, zero-shot generalize to new tools and MCP servers, navigate computer interfaces with minimal intervention, write automation scripts, and generate batches of actions. On coding, the model can diagnose and fix bugs in complex codebases, implement features, and support planning mode and subagent delegation. For multimodal tasks, it handles image, video, and PDF inputs, with strengths in visual-to-code generation, descriptive captioning, and multimodal workflow execution. The model supports a 1-million-token context window with retrieval and context compaction capabilities.

On benchmarks, Muse Spark 1.1 shows significant improvement over its predecessor across Meta’s internal coding bench and, when tested against Box’s enterprise work evaluation set, delivers performance competitive with leading frontier models. Safety evaluations place it within safe margins across Chemical and Biological, Cybersecurity, and Loss of Control categories under Meta’s Advanced AI Scaling Framework.

The Meta Model API, launched in public preview alongside Muse Spark 1.1, provides developers with an OpenAI-compatible interface to integrate the model’s agentic capabilities into their applications. By offering a direct first-party API, Meta is now competing more directly against OpenAI, Anthropic and Google for developer relationships and usage revenue, while giving enterprises a hosted path to Muse Spark 1.1 without managing their own infrastructure. The release also coincides with the launch of Muse Image, a separate model Meta says advances its vision of personal superintelligence.

KEY QUOTE:

“With these improvements, Muse Spark 1.1 advances the performance-efficiency frontier. Together with this week’s launch of Muse Image, this release brings us closer to our vision of personal superintelligence: models that help you pursue your goals, create what you imagine, deepen your relationships, and take action on what you value most.”

Meta statement