Meta unveils 'Muse Code' terminal agent to challenge OpenAI’s Codex and Claude code

Meta has officially launched Muse Code, an autonomous, terminal-based developer agent designed to execute end-to-end software engineering tasks across large repositories. Powered by the newly updated Muse Spark 1.2 model, the platform directly targets market leaders like Anthropic’s Claude Code and OpenAI’s Codex by combining multi-agent parallelism with aggressive pricing.

Autonomous Multi-Agent Architecture

Unlike traditional auto-complete coding assistants, Muse Code operates as a self-directed environment capable of planning repository-wide changes, writing code, executing terminal commands, and validating output with minimal developer steering.

• Persistent Background Sub-Agents: Instead of spawning fresh models per prompt, Muse Code maintains active background sub-agents that preserve project context throughout a session.

• Parallel Execution Workspaces: The tool can split large assignments across multiple isolated workspaces simultaneously, allowing developers to build multiple features at once without file collision risks.

• Crash-Resistant Event Logging: Every model call, terminal command, and code modification is saved to a local, append-only event log. If an operation or session fails mid-task, the agent resumes from the exact state of execution rather than starting over.

Under-Cutting Rivals on Pricing

Cost is Meta's central wedge to drive rapid enterprise and individual adoption.

• Standard Pay-As-You-Go Tier: Priced at $1.25 per million input tokens and $4.25 per million output tokens.

• Contributor Discount Tier: Offers a heavily reduced rate of $0.10 per million input tokens and $0.20 per million output tokens in exchange for opt-in feedback to train future iterations.

This pricing strategy significantly undercuts competitor rates, where premium models like Anthropic's Sonnet family run upwards of $3 per million input tokens and $15 per million output tokens.

Escalating the Enterprise Developer Race

The release underlines a broader industry shift away from standalone conversational chatbots toward agentic workflows that replace integrated development environments (IDEs) and automate heavy engineering overhead. By pairing long-horizon task completion with low-cost inference, Meta is positioning Muse Code to capture enterprise accounts as companies race to deploy autonomous coding across their software teams.