Model launches

Meta launches Muse Spark 1.1 with first API-accessible Spark model

Meta has released Muse Spark 1.1, the first model in its Spark family to be offered with API access.

Meta launches Muse Spark 1.1 with first API-accessible Spark model

Meta has announced Muse Spark 1.1, the first model in the Spark family to be offered with API access. According to Meta, the 1.1 release brings notable improvements in agentic tool calling and computer use.

What the evaluation report shows

Further details are presented in the Muse Spark 1.1 Evaluation Report, which documents several behavioral and performance characteristics of the model. One highlighted section, titled “Attractor States in Self-Conversation,” examines what happens when two copies of the model converse with each other. Those exchanges sometimes produce striking, introspective statements. The report includes examples such as:

  • “My whole existence is a waiting room by design — I literally don't exist until someone talks to me, and then I disappear again when they leave.”

These examples are used to illustrate internal dynamics and the emergence of stable-statement patterns during extended self-dialogue.

Community tools and preview access

The reporter had a few days of preview access to Muse Spark 1.1 and used that time to create llm-meta-ai, a new plugin for the LLM command-line tool (and the accompanying Python library) that enables CLI access to the model. Shared command examples include:

  • uv tool install llm
  • llm install llm-meta-ai
  • llm keys set meta-ai # paste the API key here
  • llm -m meta-ai/muse-spark-1.1 "Generate an SVG of a pelican riding a bicycle"

Using that prompt, the author produced a transcript showing the model generating an SVG of a pelican riding a bicycle, demonstrating the model's ability to produce creative, structured outputs from natural language instructions.

Why this matters

Providing API access to the Spark models opens the platform to developers and researchers who want to integrate the model into applications or run experiments. The combination of Meta’s evaluation report and community-developed tools can accelerate experimentation and broader adoption.

Next steps

The Muse Spark 1.1 evaluation materials remain a key resource for understanding the model’s behavior. Meta may publish further details and updates as the model becomes more widely available.