Muse Code Debuts, Challenging Anthropic and OpenAI
Muse Code is Meta's terminal AI coding agent; pay-as-you-go pricing and a contributor tier could lower developer costs and pressure pricier rivals.
KEY TAKEAWAYS
- Muse Code launched in beta as a terminal AI coding agent.
- Standard API prices are $1.25 input, $0.15 cached and $4.25 output per million tokens.
- Co-trained Muse Spark 1.2 supports a 1 million-token context window for long workflows.
HIGH POTENTIAL TRADES SENT DIRECTLY TO YOUR INBOX
Add your email to receive our free daily newsletter. No spam, unsubscribe anytime.
Meta Platforms launched Muse Code in beta on Aug. 5, 2026, a terminal-based AI coding agent powered by Muse Spark 1.2. The tool runs through the Meta Model API on a pay-as-you-go basis and is positioned to compete with Anthropic and OpenAI.
Terminal-Based AI Coding Agent for Developers
Muse Code is Meta’s first AI coding agent, available in beta for macOS and Linux. It installs with a single command and operates entirely through the terminal and the Meta Model API, without a dedicated graphical interface. For example, users can install it via the command: curl -fsSL https://dev.meta.ai/install.sh | bash.
Designed for complex software engineering across large codebases, Muse Code can plan changes, write code, validate results, and assist with debugging. It coordinates multiple persistent subagents per task, including parallel workers that implement changes and background reviewers that validate outcomes. These asynchronous agents remain active throughout a session, making every step observable to the developer.
The agent pairs with Muse Spark 1.2, a coding-focused update to the Muse Spark model family. This model features significantly increased training compute on coding tasks and broader training diversity, which Meta says improves code generation, complex debugging, codebase understanding, and long-running workflows. Muse Spark 1.2 supports a 1 million-token context window, enabling extended sessions and handling large repositories within a single workflow.
Meta describes Muse Code and Muse Spark 1.2 as co-trained, a process that reduces retries and improves tool use compared with treating the model as an external wrapper.
Access, Pricing, and Competitive Positioning
Access to Muse Code is provided through the Meta Model API, which is in public preview with expanded global reach compared to earlier US-only access. The standard pay-as-you-go pricing for Muse Spark 1.2 API usage is $1.25 per million input tokens, $0.15 per million cached input tokens, and $4.25 per million output tokens. Meta commits that prompts and completions on the standard tier are not used to train its models.
A lower-cost contributor tier charges $0.10 per million input tokens, $0.002 per million cached input tokens, and $0.20 per million output tokens in exchange for allowing Meta to use contributor data to improve its models.
Meta and coverage frame Muse Code and Muse Spark 1.2 as a cheaper alternative to Anthropic’s Claude Code and OpenAI’s Codex and related GPT-based coding features. The rollout is led by Meta Superintelligence Labs under AI chief Alexandr Wang. Meta describes these releases as steps toward larger, more capable frontier models and what it calls personal superintelligence.
Meta has not provided revenue, user-adoption, or margin guidance specifically tied to Muse Code or Muse Spark 1.2 in the available materials.
