Meta has launched Muse Code, a terminal-based AI coding agent designed for large software projects, with pricing that undercuts competitors like Claude Code and OpenAI Codex.
TL;DR
- Meta introduces Muse Code, an AI coding agent for comprehensive software engineering tasks.
- The agent features multi-agent task coordination and persistent activity logs for complex projects.
- Muse Code's pricing model may attract cost-conscious developers, but real-world performance remains unproven.
What happened
Meta released Muse Code in beta on 5 August, an AI coding agent that operates within a developer's terminal. The agent is powered by Meta's Muse Spark 1.2 model and is designed to handle large software projects by planning changes, editing multiple files, and testing software functionality.
Muse Code can run several sub-agents simultaneously to manage complex tasks and maintain a persistent activity log to resume interrupted sessions. Meta trained the model and agent together to optimize performance for software engineering tasks.
The standard Muse Code tier is priced at $1.25 per million input tokens and $4.25 per million output tokens, with a discounted contributor tier at $0.20 per million output tokens for users providing feedback.
Why it matters
Muse Code represents Meta's direct challenge to established coding agents like Claude Code and Codex, with a focus on enterprise-level software development tasks.
The pay-as-you-go pricing model may attract cost-conscious developers, particularly in regions where AI services are expensive relative to local currencies.
However, the real measure of Muse Code's success will be its reliability in handling undocumented systems, messy dependencies, and company-specific development practices.
Key facts
- Muse Code was released in beta on 5 August.
- The agent is powered by Meta's Muse Spark 1.2 model.
- Standard tier pricing: $1.25 per million input tokens, $4.25 per million output tokens.
- Contributor tier pricing: $0.20 per million output tokens with feedback requirements.
- Meta evaluated Muse Code across 89 terminal tasks and 113 software-engineering tasks from 91 repositories.
- Supported languages include TypeScript, Go, Python, JavaScript, and Rust.
- Muse Code can coordinate multiple sub-agents for complex tasks.
- The agent maintains a persistent activity log for session resumption.
Context
Meta's entry into the coding agent market follows the expansion of competitors like Anthropic's Claude Code and OpenAI's Codex, which have integrated coding capabilities into their platforms.
The rise of AI coding agents reflects a broader trend in the AI industry towards more sophisticated tools that assist developers in complex software engineering tasks.
As AI models become more capable, the competition among tech giants to capture the developer market is intensifying, with pricing and performance as key differentiators.
