Meta launches Muse Code to compete with Codex and Claude Code

Meta has launched Muse Code, a beta terminal coding agent for macOS and Linux, alongside a new coding-focused model called Muse Spark 1.2. The company says the agent can plan changes, write code and validate results across large repositories.

The release is Meta’s clearest move into the same coding-agent market now led by OpenAI Codex and Anthropic Claude Code. Muse Code uses persistent background agents that stay active through a session instead of being spawned only for one subtask. Meta says that design reduces repeated context gathering and helps with long, multi-step software tasks.

Meta is also emphasizing reliability mechanics rather than just model scores. Muse Code keeps a local event log of model calls, tool runs, approvals and edits, which is meant to make sessions restart-safe and replayable after failures. Muse Spark 1.2 was co-trained with the agent and tuned for code generation, debugging, codebase understanding and long-horizon workflows.

For readers, the signal is that coding agents are becoming a strategic platform layer, not a side feature inside chatbots. Meta now has its own terminal product, its own coding model update and a cost-focused pitch for developers comparing agentic coding tools.