Meta has launched Muse Code, its first AI agent designed for software development, now available in beta. The tool goes up against Anthropic's Claude Code and OpenAI's Codex, with Meta betting that closer model integration and more competitive pricing will win developers over.
Muse Code helps developers plan changes, write code, run it, and verify that it works from a single interface, coordinating multiple specialized agents in the process. It runs on the new Muse Spark 1.2 model, which Meta built alongside the agent to improve programming performance compared with the previous version. The launch is the first major announcement from Alexandr Wang, who was appointed to lead Meta Superintelligence Labs in June 2025.
Competing on price, not raw performance
Meta is offering Muse Code on a pay-as-you-go basis, with pricing comparable to Muse Spark 1.1. There's also a "Contributor" plan at more than ten times lower cost, in exchange for permission to use user interactions to improve the model.
Rather than trying to immediately outperform Anthropic's Claude Code or OpenAI's Codex on raw performance, Meta built infrastructure designed specifically for programming projects. Developers can still use other models if they choose. Wang believes the best performance will come from pairing Muse Code with Muse Spark, since the two were developed together.
Zero data retention for enterprises
Meta also announced a "zero data retention" option for companies that want their information neither stored nor used for model training. The feature addresses growing demand among business customers for privacy, and fits Meta's effort to strengthen its enterprise offering as competition in AI coding assistants intensifies.
Why this matters for IT and Development
The pricing structure is the main thing to watch. The Contributor plan makes an AI coding agent accessible at a fraction of the usual cost, but the trade-off is that your interactions train Meta's model - a real concern for teams working on proprietary code. For sensitive work, the zero data retention option avoids that problem entirely.
Muse Code also lets teams plug in other models, so adopting the tool doesn't mean locking into a single vendor. For those tracking AI for IT & Development, it's a new option worth testing against Claude Code and Codex. Developers who want structured training on AI coding tools and automation can work through the AI Learning Path for Software Developers.
Your membership also unlocks:
