Meta's Muse Code cuts costs—but wants your source code in return

Meta has launched Muse Code, an AI-powered coding agent designed to handle complex software engineering tasks—from planning and writing code to testing and validation. Built on the company's recently updated Muse Spark 1.2 model, the tool promises competitive pricing that undercuts rivals like Anthropic's Claude Code and OpenAI's offerings.

The pricing hook—and the data catch#

The catch lies in how Meta's pricing works. While standard pay-as-you-go rates ($1.25 per million input tokens, $4.25 per million output tokens) position Muse Code as a budget option, the real incentive sits in the "contributor tier." According to Meta's chief AI officer, this tier costs more than 10 times less than standard pricing—but only if users agree to let Meta use their code and interactions to improve its models.

That data-sharing requirement includes not just source code but developer prompts, agent responses, and manual corrections made during coding sessions. The implication: competitors or bad actors could potentially gain insight into a company's unique approaches and proprietary techniques.

Why engineering leaders are skeptical#

Few software executives seem tempted by the discount. Ken Ringdahl, CTO at Emburse and advisor to Snyk, frames the trade-off bluntly: "Security, data privacy, and data security is paramount, and it's not something we will trade off for a lower-cost option." He adds, "There isn't IP that's more important than your source code itself. I won't gamble with my IP."

Yunhao Jiao, CEO of TestSprite, raises a related concern: even if Meta handles the data responsibly, there's no guarantee it stays secure long-term. If a future breach or leak occurs, competitors could gain access to years of proprietary development patterns embedded in those training datasets.

Where Muse Code might fit#

The skepticism doesn't mean the tool has no use case. Developers and companies working on open-source projects, personal experiments, or disposable prototypes may find the contributor tier worthwhile. Michał Piszczek, CTO at Archdesk, envisions a hybrid approach: startups using cheaper, data-sharing options for public codebases while reserving zero-data-retention tiers for core products.

Meta has indicated it's exploring zero-data-retention options—a setup where the company wouldn't retain or train on developer data—though no timeline or pricing has been announced. Muse Code is currently available in beta.