Meta enters the coding-agent race: Muse Code ships in beta with Muse Spark 1.2

Meta announced Muse Code on 5 August 2026, describing it as a terminal coding agent that "takes on complex software engineering tasks across large repositories: planning changes, writing code, and validating the results". It ships in beta for macOS and Linux, installed with a single shell command, and runs on Muse Spark 1.2 — a coding-focused update to the Muse Spark 1.1 model released in July. The design choice worth noting is how the agent handles parallel work. Meta says its background agents "remain active throughout each session, rather than being spawned for individual tasks", with a local event log that makes a session replay-exact and restart-safe — the agent picks up where it stopped instead of losing the thread after a crash. Three bundled skills come with it: /plan, which gates a proposed plan behind human approval, /grill for stress-testing work already done, and /goal for driving an objective to completion. Meta's own post does not publish pricing or benchmark tables in text form. Trade coverage of the launch reports a one-million-token context window, pay-as-you-go rates of $1.25 per million input tokens and $4.25 per million output tokens — the same as Muse Spark 1.1 — a "contributor" tier more than ten times cheaper for developers who opt in to helping improve the model, and a score of 59% on the DeepSWE 1.1 benchmark. wujec.ai treats those figures as press reporting rather than vendor documentation until Meta states them itself. The strategic point is simpler than the numbers. Coding agents have been the most commercially defensible product category in generative AI, and until this week Meta had no entry in it. Neither Muse Code nor the Muse Spark model line has a profile in this catalogue yet — a gap on our side, and one we are adding to the queue.