Muse Spark 1.1, announced on 9th July 2026, is the first Spark model to offer an API, and Meta says it delivers significant improvements in agentic tool calling and computer use. Simon Willison linked to the Muse Spark 1.1 Evaluation Report and shared notes from a short preview of the model.

What is Muse Spark 1.1?

Muse Spark 1.1 is an updated Spark-family model released 9th July 2026 that, for the first time in the Spark line, exposes an API and is claimed by Meta to improve agentic tool calling and computer use. The release follows the original Muse Spark from April, and Meta published an Evaluation Report with further detail.

Meta framed agentic tool calling and computer use as the areas with "significant improvements" in Muse Spark 1.1. The Evaluation Report is the primary place Meta put those details, and Willison highlighted a section titled "Attractor States in Self-Conversation" where two copies of the model talking to each other produced striking lines.

One example line from that self-conversation reads: "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." That quote illustrates behavioural observations recorded in the Evaluation Report and shared in Willison's link post.

How can developers try Muse Spark 1.1?

Developers can access Muse Spark 1.1 via the new API that accompanies this release; Simon Willison described a short preview and demonstrated CLI access through a new plugin called llm-meta-ai. He published the exact commands he used to install and call the model from the command line.

The sequence Willison provided for trying the model was:

uv tool install llm llm install llm-meta-ai llm keys set meta-ai

paste API key here

llm -m meta-ai/muse-spark-1.1 "Generate an SVG of a pelican riding a bicycle"

Willison said a few days of preview access let him build llm-meta-ai, a plugin offering CLI and Python library access to the model. He included a transcript of the pelican example in the post.

How does this compare to the prior Muse Spark release?

Muse Spark 1.1 follows Muse Spark, which appeared in April 2026, and represents a shift in distribution by being the first Spark model to expose an API. The core claims for 1.1 are improved agentic tool calling and improved computer use, and Meta published an Evaluation Report to document those changes.

Willison presented 1.1 as a point release that adds access and behavioural tweaks rather than a ground-up replacement. The Evaluation Report and Willison's preview notes are the only sources he linked for readers to explore the technical details and examples.

Why it matters

An API changes who can integrate Spark models. Exposing Muse Spark 1.1 via an API makes it easier for developers to call the model from command-line tools and libraries, as Willison demonstrated with llm-meta-ai. Improved agentic tool calling and computer use are the specific behavioural areas Meta highlights, which matter for workflows that rely on models interacting with external tools or performing multi-step tasks.

What to watch

Read the Muse Spark 1.1 Evaluation Report for the technical details Meta published, and watch for wider availability beyond the preview access Willison had. The next signals to track are expanded developer access to the API and concrete examples of the claimed tool-calling and computer-use improvements in real integrations.