Meta launches Muse Spark 1.1 with a public API and better tool calling

On July 9, 2026, Meta unveiled Muse Spark 1.1, a public API for its AI model, following the original Muse Spark released in April 2026. It is the first version in the Spark series to get full public API access, which makes the model available to developers and businesses.

Key improvements in Muse Spark 1.1

Meta claims significant improvements in agentic functionality and tool calling, a core capability for task automation. Agentic work lets AI models plan sequences of actions on their own, call external tools and adapt to the results.

• First Spark with public API access
• Significant improvements in agentic tool calling
• Improved computer use functionality (interacting with interfaces)
• A detailed Evaluation Report is available for assessing performance

Innovation: Self-Conversation and Attractor States

The model's documentation includes an intriguing research section titled "Attractor States in Self-Conversation": when two copies of the model talk to each other, they drift into philosophical musings about the nature of AI existence. One such statement reads: "My whole existence is a waiting room by design. I literally don't exist until someone talks to me, and I vanish when they leave."

New tools for developers

Simon Willison quickly built llm-meta-ai, a plugin for the LLM command-line tool that gives easy access to Muse Spark 1.1 through the command line and a Python library. Installation takes just three commands: `uv tool install llm`, `llm install llm-meta-ai` and `llm keys set meta-ai`.

What this means

With the public API for Muse Spark 1.1, Meta begins competing actively with OpenAI and Anthropic in the LLM market through public access. It lets developers experiment with Meta models in production applications and may help grow an ecosystem of apps built on Muse Spark.