Introducing Muse Code and Muse Spark 1.2

Meta released Muse Spark 1.2 along with its own coding agent called Muse Code. This update focuses on long-sequence agentic tool calling and improves code generation, complex debugging, codebase understanding, and end-to-end developer workflows.

To achieve this performance, Meta significantly increased training compute for coding tasks and expanded training environment diversity. Muse Spark 1.2 and Muse Code were co-trained using rejection sampled harness trajectories, recipe optimizations for goals, compaction, and subagents, and the integration of the Muse Code toolset. The model underwent extensive training on long-horizon coding tasks such as whole-repository generation, large end-to-end projects, and auto-research.

The model is available under two distinct pricing options. The standard model ID, muse-spark-1.2, costs $1.25 per million input tokens and $4.25 per million output tokens. A discounted version named muse-spark-1.2-contributor is available for $0.10 per million input tokens and $0.20 per million output tokens, provided users agree to let Meta use their data to improve products.