Meta Releases Muse Spark 1.2 Frontier Coding Model and Muse Code Agent

Meta has expanded its artificial intelligence tooling with the release of Muse Spark 1.2 and Muse Code, entering a competitive market for developer assistance and terminal coding agents.

Muse Code Agent and Architecture

Meta introduced Muse Code as a terminal coding agent currently available in beta. The software is designed to handle software engineering tasks across large code repositories by planning changes, writing code, and validating results.

Unlike harnesses that spawn helper agents fresh for each task, Muse Code uses persistent async background agents that remain active throughout a session to avoid redundant information gathering. The system also maintains a local event log tracking model calls, tool runs, approvals, and edits. This single source of truth makes the runtime restart-safe, allowing the agent to resume precisely where it left off after a crash.

Muse Spark 1.2 Model Training

Underlying the new agent is Muse Spark 1.2, a coding-focused update to Meta's model line. Meta co-trained the model with Muse Code to ensure optimal performance when paired together. The training process utilized a self-improvement loop where Muse Spark 1.1 generated challenging coding environments and instruction-following templates, grading candidate solutions to produce a scalable training dataset for the newer version.

Availability and Pricing

Muse Spark 1.2 is available through the Meta Model API with expanded global access, while Muse Code can be installed on macOS or Linux via a terminal script. Standard API pricing is set at $1.25 per million input tokens and $4.25 per million output tokens. Meta also offers a lower-cost contributor tier in exchange for permission to use prompts and completions to train future models.