Meta Launches Muse Code and Muse Spark 1.2 with Coding Agent Focus

Meta shipped its own coding agent as part of demonstrating that long-sequence agentic tool calling is the most important capability of models today. Muse Spark 1.2 is a coding-focused update to Muse Spark 1.1, with improvements in code generation, complex debugging, codebase understanding, and end-to-end developer workflows. The training significantly scaled up compute on coding tasks while expanding training environment diversity, and the model retains strength in areas like general agents. Muse Spark 1.2 was co-trained with Muse Code, using rejection sampled harness trajectories and recipe optimizations for goals, compaction, and subagents, alongside integration of the Muse Code toolset. It was extensively trained on long-horizon coding tasks, including whole-repository generation, large end-to-end projects, and auto-research. Simon Willison compared the quality of an SVG pelican produced by Muse Spark 1.2, finding it a small but material improvement over the 1.1 version.