Meta Releases Muse Code Coding Agent With Muse Spark 1.2

Muse Code combines a primary agent loop with specialized background agents that remain active throughout a session. According to Meta, this design is intended to reduce repeated information gathering during long tasks. The runtime records model calls, tool runs, approvals, and edits in a local event log, which gives the system a recoverable history when work branches or fails.

The beta also exposes workflow controls for planning and review. Meta describes an approval-gated planning command, a plan stress-test command, and a goal mode that continues toward a specified result. Those controls are useful signs of an agent designed for repository-scale work, but they do not eliminate the need for human review of proposed changes, tool permissions, tests, and security-sensitive operations.

Muse Spark 1.2 is a coding-focused update to Muse Spark 1.1. Meta says it increased coding-task training compute, diversified training environments, and co-trained the model with the Muse Code harness. The company also says the model was trained for long-horizon work such as whole-repository generation and large end-to-end projects, using planning, goal conditioning, and context compaction to sustain progress. These are vendor-reported design and performance claims; independent evaluation will be needed to establish how reliably the beta handles real production repositories.

CNBC reports that developers can access Muse Code through a pay-as-you-go option using pricing similar to Muse Spark 1.1: $1.25 per million input tokens and $4.25 per million output tokens. For engineering teams, the practical question is not only benchmark performance but whether the event log, approval gates, and validation loop make long-running agent work observable and reversible enough for controlled use.