Meta Superintelligence Labs released Muse Code in beta on August 5, 2026, pairing it with a new coding-focused model, Muse Spark 1.2, in what the company framed as a single coordinated product rather than a model upgrade followed by tooling. The release is a terminal coding agent powered by Muse Spark 1.2, the company's newest model. The launch puts Meta in direct competition with Anthropic's Claude Code, OpenAI's Codex, and the growing field of agentic coding harnesses that have rapidly become the primary way many professional developers ship software. Chief Executive Mark Zuckerberg announced the release on X, writing that it is "a terminal coding agent that takes on complete software engineering tasks across large repos: planning changes, writing code, validating the results."
The architecture distinguishes Muse Code from most rivals on a technical dimension that matters to enterprise buyers. The agent can automatically break a larger project into separate tasks handled by parallel subagents, with each write-capable subagent operating inside its own isolated Git worktree, allowing multiple agents to modify code simultaneously without interfering with one another or altering the developer's main working copy. Sessions can persist after a laptop closes or a process crashes, and the "muse resume" command can reconstruct the previous session from its log. Every subagent spawned, every tool call, every steer and cancel, is observable and replayable through an event log, a transparency feature Meta is marketing to teams that need audit trails. In a stress test Meta published alongside the launch, the model optimized GPU kernels over more than 1,000 tool calls and up to 24 hours of continuous work.
The pricing structure may carry more competitive weight than any benchmark. Standard Meta Model API pay-as-you-go pricing is $0.15 per million cached input tokens, $1.25 per million input tokens, and $4.25 per million output tokens. Meta also offers a "contributor" tier at significantly lower cost, in exchange for using user prompts and completions to train its models, a strategy that lets Meta acquire real-world coding data that is scarce and superior to scraped or synthetic data because it is checkable. That two-tier structure lands against a field where GitHub Copilot's Max tier runs to $100 a month and Claude Code's maximum plan reaches $200 a month. The financial viability of these tools remains tenuous, evidenced by GitHub's recent decision to halt new Copilot sign-ups as usage levels surged beyond sustainable limits.
Skeptics have raised legitimate concerns about the beta's enterprise readiness. One comparison published days after the launch argued that GitHub Copilot is the defensible choice for GitHub-centric organizations because its product documentation, billing, IDE integrations, security controls, and support channels can be audited in ways Muse Code's beta cannot yet match. The model itself, while co-trained with the Muse Code harness, is described by Meta's own developers as a moderate improvement over Muse Spark 1.1, optimized for the work that coding agents get handed most, such as multi-file refactors, long debugging sessions, and tasks that run well past a single prompt. Muse Spark 1.2 is the third release in four months from Meta Superintelligence Labs, a pace that signals ambition but also the kind of rapid iteration that enterprise procurement officers treat with caution.
Any competitor still pricing agentic coding access above $2 per million input tokens now has to explain why to every engineering manager who read Tuesday's announcement.