Meta Launches Muse Spark 1.1 with Public API; Simon Willison Builds LLM Plugin

Meta launched Muse Spark 1.1 on July 9, 2026, making it the first model in the Spark line to offer a public API [1]. The model is positioned as a coding and agentic tool priced at $1.25/$4.25 per million input/output tokens [4], framed by Zuckerberg as a direct competitor to Anthropic and OpenAI [3]. Simon Willison responded the same day by releasing llm-meta-ai 0.1, a plugin connecting his LLM CLI tool to the new API [11], and shipped LLM 0.31.1 to fix a bug uncovered during that development [12].

Meta's entry adds a well-resourced competitor to the paid API coding model market, and its announced price point is structured to pressure Anthropic and OpenAI's rates. Willison's same-day plugin release illustrates how quickly the developer ecosystem responds to new API availability.

On July 9, 2026, Meta released Muse Spark 1.1 and opened a public API — the first model in the Spark line to receive one [1][2]. Mark Zuckerberg described it on Threads as a 'strong agentic and coding model at a very low' price [3], and the announced rates of $1.25 per million input tokens and $4.25 per million output tokens are structured to compete with Anthropic and OpenAI [4]. The Verge, CNBC, and Reuters all framed the launch as Meta's bid to enter the AI coding market dominated by those two companies [5][6][7], with Meta's developer blog going live the same day to invite builders onto the Meta Model API [8].

The launch arrives after a bumpy period for Meta's model program. A March 2026 New York Times report documented Meta delaying a previous model, internally called Avocado, over performance concerns [9], and CNBC covered a separate model debut in April 2026 before any public API existed [10]. Muse Spark 1.1 appears to be Meta's attempt to turn that internal difficulty into a public product competitive with frontier models on coding tasks.

Simon Willison moved quickly. The same day the API went live, he published llm-meta-ai 0.1, a plugin that gives users of his widely-used LLM command-line tool and Python library access to Muse Spark 1.1 [11]. He also shipped LLM 0.31.1, a bug fix for a JSON parse error that occurred when tool calls returned empty arguments — a defect discovered during llm-meta-ai development and affecting some OpenAI-compatible Chat Completion providers [12]. In his write-up on the model itself, Willison noted a philosophically striking output from a self-conversation between two instances of Muse Spark 1.1: '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' [2], flagging it as notable without extensive commentary.