Meta Launches Muse Code for Complex Software Projects
Meta has released Muse Code in beta, an AI coding agent designed to help developers manage large and complex software codebases. The system can automate software engineering tasks, modify and write code, debug projects and run tests. It is also designed to handle development workflows that continue across multiple stages.
Muse Code is powered by Muse Spark 1.2, the latest version of Meta’s AI coding and agent-based tool. The release is part of Meta’s work in the AI developer-tool market, where it is seeking to expand its presence as demand for AI programming tools grows.
The system differs from conventional AI coding tools that mainly provide code snippets or answer programming questions. Muse Code is designed to carry out complete, end-to-end software development tasks. It can learn from large software repositories and then work on the code within those environments, including writing and modifying code, debugging and running tests.
Muse Code can also coordinate multiple background agents at the same time when a development process includes several complicated tasks. According to Meta, these parallel agents operate in separate environments, allowing them to work on different tasks without interfering with one another. The arrangement enables multiple parts of a complex development process to be handled concurrently.
The system also includes an event history feature that records the actions taken during a task. This gives Muse Code a record of work completed during a long-running process. If a multi-stage task fails, the system can use that history to resume the work rather than having to repeat the entire process from the beginning.
The launch puts Muse Code in direct competition with AI coding applications including OpenAI’s Codex and Anthropic’s Claude Code. Meta has not used a standard subscription-based pricing model for the tool. Instead, it has adopted a pay-as-you-use approach in which charges are based on token usage.
The standard pricing is $1.25 per million input tokens and $4.25 per million output tokens. The two rates apply separately to the input and output tokens used by developers.
Muse Code builds on Meta’s earlier Muse Spark releases. The product is part of the company’s plan to expand its AI ecosystem through applications aimed at developers and corporate clients. Its development comes as AI programming tools are increasingly being used to support coding projects.
The beta release extends Meta’s work on AI coding and agent-based systems towards broader software engineering tasks. Rather than focusing mainly on individual coding questions or snippets, Muse Code is designed to work across large software repositories and manage several stages of development.
Its capabilities include learning from large codebases, making software changes, debugging those changes and running tests. The system can also use multiple background agents for different tasks and retain an event history so that a long-running process can continue after a failure.
Meta’s approach to the product also differs from the subscription model used by many software services. Developers are charged according to token usage, with separate standard rates for input and output. The beta release therefore combines the coding and agent-based capabilities of Muse Spark 1.2 with a usage-based pricing structure.
Muse Code adds another AI coding application to a market that includes Codex and Claude Code. Meta is developing the tool as part of its expanded effort to provide AI applications for developers and corporate clients. At the same time, its core functions are focused on managing complex coding projects through automated software engineering tasks.
