Meta launches coding-focused Muse Spark 1.2 for long and complex software projects

Meta has launched Muse Code, its newest AI coding agent built to help developers work on large and complex software projects with minimal manual effort. Powered by the newly released Muse Spark 1.2 model, the tool is designed to plan code changes, write new code and verify results across large code repositories. The launch marks Meta's biggest push yet into AI-powered software development, as competition intensifies with Anthropic's Claude Code, OpenAI's Codex, and Google's growing portfolio of coding tools.

More than just an AI coding assistant

Unlike traditional AI coding assistants that respond to one prompt at a time, Muse Code works as an autonomous software engineering agent. Meta says the system creates multiple persistent subagents that work together on a single task. These background agents remain active throughout a session, allowing them to retain context instead of repeatedly analysing the same codebase. The company says this approach helps developers tackle long-running engineering tasks faster while reducing repetitive work.

Built for large software projects

At the heart of Muse Code is Muse Spark 1.2, an upgraded AI model trained specifically for coding. Meta says it significantly increased training compute and expanded the range of coding environments used during training. The model was also co-trained alongside Muse Code to improve how the two systems work together. According to Meta, Spark 1.2 is optimised for long-horizon tasks such as generating complete repositories, building end-to-end applications and managing large enterprise codebases. Another standout feature is crash recovery. Muse Code keeps a detailed event log of every AI action, tool execution and model call, allowing the agent to resume exactly where it stopped if interrupted.

Meta joins the AI coding race

The launch comes as AI companies increasingly compete to build autonomous software engineering agents rather than simple coding assistants. Meta says Muse Spark 1.2 is available globally through Muse Code and the Meta Model API starting today. The company also confirmed that even more powerful coding models and new agent capabilities are already under development. With developers increasingly looking for AI that can manage entire projects instead of generating isolated snippets of code, Muse Code signals Meta's ambition to become a major player in enterprise AI development tools.