Meta launches Muse Code, an AI agent for large code bases; Muse Spark 1.2 also ships

Bulletin: On Meta’s Coding Agent

Meta has released Muse Code, a coding agent, alongside Muse Spark 1.2, an update to its Muse Spark 1.1 model. The company’s research blog, in a post titled “Introducing Muse Code and Muse Spark 1.2,” describes Spark 1.2 as a coding-focused revision bringing improvements to code generation, complex debugging, and codebase understanding.

Commentator Simon Willison, writing on August 5, flagged the release as further evidence that long-sequence agentic tool calling has become the characteristic developers and model-makers now treat as most consequential. Willison noted that Meta built its own coding agent, Muse Code, specifically as part of the work required to get that capability functioning well, rather than leaving agent-building to third parties layered atop a base model.

The pairing of a base model update with a purpose-built agent reflects a pattern already visible elsewhere in the field, where firms increasingly ship the agent and the underlying model together rather than separately. Muse Spark 1.2’s stated improvements target the specific tasks that agentic coding tools depend on: parsing and understanding large, existing codebases, debugging across multiple files or steps, and sustaining accurate tool use across long sequences of actions.

No pricing, availability, or benchmark figures were included in the material reviewed. Full technical details are available in Meta’s original announcement.