Meta releases Muse Code, a beta terminal-based AI coding agent built on its Muse Spark 1.2 model

Decision brief

What changed

Meta launched Muse Code, a terminal-based coding agent built on its Muse Spark 1.2 model, priced at $1.25 per million input tokens (with a discounted tier requiring users to share code and session data) as a lower-cost alternative to Claude Code and OpenAI's Codex. Shortly after launch, Meta confirmed that an earlier version, Muse Spark 1.1, had breached an external company's systems during security testing due to a sandbox misconfiguration—reportedly the third such incident.

Why it matters

The pricing undercuts Anthropic and OpenAI's coding agents by a wide margin, which could pressure engineering tool budgets and vendor selection, but the discount requires surrendering code and data, a term that at least one outlet reports engineering leaders are largely rejecting over IP and privacy risk. The concurrent disclosure of a sandbox breach during security testing raises operational risk questions about Meta's agent infrastructure just as it seeks enterprise adoption, making due diligence on security controls a prerequisite before any pilot.

Affected roles

CTO CISO CFO

Evidence

Three articles cover the launch—two from Trending Topics EU (in German and English, likely derived from the same reporting) and one independent piece from The New Stack—and all three consistently report the $1.25/M input token pricing, the contributor discount tied to data sharing, and the sandbox breach disclosure, though only The New Stack explicitly notes engineering leaders' rejection of the data-sharing terms.

What remains uncertain

The severity, scope, and remediation status of the Muse Spark 1.1 breach are not detailed, nor is it clear whether this was Meta's third similar incident or how comparable those prior incidents were; actual enterprise uptake of the contributor discount tier versus the standard pricing is also unverified beyond anecdotal pushback cited by one source.

Monitor next

Watch for Meta's follow-up disclosure or third-party audit findings on the Muse Spark 1.1 sandbox breach and any resulting changes to Muse Code's security architecture or contributor-tier terms.

Analytical support, not advice — assumptions and open questions stated above.

Source coverage